Libro ilustrate dstevens tcpip

TCP/IP Illustrated
Oleg V. Titov, CTD IFMO RU

Table of Contents
TCP/IP Illustrated . . . . . . . . . . . . . . . . . 1
TCP/IP Illustrated, Volume 1 . . . . . . . . . . . . . . 1
The Protocols . . . . . . . . . . . . . . . . . 1
W. Richard Stevens . . . . . . . . . . . . . . . . 1
Contents . . . . . . . . . . . . . . . . . . 1
Preface . . . . . . . . . . . . . . . . . . . . 8
Preface . . . . . . . . . . . . . . . . . . . 8
Introduction . . . . . . . . . . . . . . . . . 8
Organization of the Book . . . . . . . . . . . . . . 8
Readers . . . . . . . . . . . . . . . . . . 9
Systems Used for Testing . . . . . . . . . . . . . . 10
Typographical Conventions . . . . . . . . . . . . . . 11
Acknowledgments . . . . . . . . . . . . . . . . 11
Chapter 1. Introduction . . . . . . . . . . . . . . . . 13
Introduction . . . . . . . . . . . . . . . . . . 13
1.1 Introduction . . . . . . . . . . . . . . . . 13
1.2 Layering . . . . . . . . . . . . . . . . . 13
1.3 TCP/IP Layering . . . . . . . . . . . . . . . 17
1.4 Internet Addresses . . . . . . . . . . . . . . . 18
1.5 The Domain Name System . . . . . . . . . . . . . 19
1.6 Encapsulation . . . . . . . . . . . . . . . . 20
1.7 Demultiplexing . . . . . . . . . . . . . . . 21
1.8 Client-Server Model . . . . . . . . . . . . . . 22
1.9 Port Numbers . . . . . . . . . . . . . . . . 23
Reserved Ports . . . . . . . . . . . . . . . . 23
1.10 Standardization Process . . . . . . . . . . . . . 24
1.11 RFCs . . . . . . . . . . . . . . . . . 24
1.12 Standard, Simple Services . . . . . . . . . . . . . 25
1.13 The Internet . . . . . . . . . . . . . . . . 26
1.14 Implementations . . . . . . . . . . . . . . . 26
1.15 Application Programming Interfaces . . . . . . . . . . . 27
1.16 Test Network . . . . . . . . . . . . . . . . 28
1.17 Summary . . . . . . . . . . . . . . . . . 28
Exercises . . . . . . . . . . . . . . . . . 29
Chapter 2. Link Layer . . . . . . . . . . . . . . . . 30
Link Layer . . . . . . . . . . . . . . . . . . . 30
2.1 Introduction . . . . . . . . . . . . . . . . 30
2.2 Ethernet and IEEE 802 Encapsulation . . . . . . . . . . . 30
2.3 Trailer Encapsulation . . . . . . . . . . . . . . 32
2.4 SLIP: Serial Line IP . . . . . . . . . . . . . . 33
2.5 Compressed SLIP . . . . . . . . . . . . . . . 34
2.6 PPP: Point-to-Point Protocol . . . . . . . . . . . . . 34
2.7 Loopback Interface . . . . . . . . . . . . . . . 36
2.8 MTU . . . . . . . . . . . . . . . . . . 38
2.9 Path MTU . . . . . . . . . . . . . . . . . 38
2.10 Serial Line Throughput Calculations . . . . . . . . . . . 39
2.11 Summary . . . . . . . . . . . . . . . . . 40
Exercises . . . . . . . . . . . . . . . . . 40

i

Chapter 3. IP: Internet Protocol . . . . . . . . . . . . . . 41
.
IP: Internet Protocol . . . . . . . . . . . . . . . . . 41
.
3.1 Introduction . . . . . . . . . . . . . . . . . 41
.
3.2 IP Header . . . . . . . . . . . . . . . . . 41
.
3.3 IP Routing . . . . . . . . . . . . . . . . . 45
.
Examples . . . . . . . . . . . . . . . . . 47
.
3.4 Subnet Addressing . . . . . . . . . . . . . . . 49
.
3.5 Subnet Mask . . . . . . . . . . . . . . . . 51
.
Example . . . . . . . . . . . . . . . . . . 51
.
3.6 Special Case IP Addresses . . . . . . . . . . . . . 52
.
3.7 A Subnet Example . . . . . . . . . . . . . . . 53
.
3.8 ifconfig Command . . . . . . . . . . . . . . 55
.
3.9 netstat Command . . . . . . . . . . . . . . 56
.
3.10 IP Futures . . . . . . . . . . . . . . . . . 56
.
3.11 Summary . . . . . . . . . . . . . . . . . 57
.
Exercises . . . . . . . . . . . . . . . . . . 58
.
Chapter 4. ARP: Address Resolution Protocol . . . . . . . . . . . 59
.
ARP: Address Resolution Protocol . . . . . . . . . . . . . . 59
.
4.1 Introduction . . . . . . . . . . . . . . . . . 59
.
4.2 An Example . . . . . . . . . . . . . . . . . 59
.
4.3 ARP Cache . . . . . . . . . . . . . . . . . 60
.
4.4 ARP Packet Format . . . . . . . . . . . . . . . 61
.
4.5 ARP Examples . . . . . . . . . . . . . . . . 62
.
Normal Example . . . . . . . . . . . . . . . . 62
.
ARP Request to a Nonexistent Host . . . . . . . . . . . . 64
.
ARP Cache Timeout . . . . . . . . . . . . . . . 65
.
4.6 Proxy ARP . . . . . . . . . . . . . . . . . 65
.
4.7 Gratuitous ARP . . . . . . . . . . . . . . . . 67
.
4.8 arp Command . . . . . . . . . . . . . . . . 68
.
4.9 Summary . . . . . . . . . . . . . . . . . 68
.
Exercises . . . . . . . . . . . . . . . . . . 68
.
Chapter 5. RARP: Reverse Address Resolution Protocol . . . . . . . . . 69
.
RARP: Reverse Address Resolution Protocol . . . . . . . . . . . 69
.
5.1 Introduction . . . . . . . . . . . . . . . . . 69
.
5.2 RARP Packet Format . . . . . . . . . . . . . . . 69
.
5.3 RARP Examples . . . . . . . . . . . . . . . . 69
.
5.4 RARP Server Design . . . . . . . . . . . . . . . 72
.
RARP Servers as User Processes . . . . . . . . . . . . 72
.
Multiple RARP Servers per Network . . . . . . . . . . . 72
.
5.5 Summary . . . . . . . . . . . . . . . . . 72
.
Exercises . . . . . . . . . . . . . . . . . . 73
.
Chapter 6. ICMP: Internet Control Message Protocol . . . . . . . . . 74
.
ICMP: Internet Control Message Protocol . . . . . . . . . . . . 74
.
6.1 Introduction . . . . . . . . . . . . . . . . . 74
.
6.2 ICMP Message Types . . . . . . . . . . . . . . 74
.
6.3 ICMP Address Mask Request and Reply . . . . . . . . . . 77
.
6.4 ICMP Timestamp Request and Reply . . . . . . . . . . . 78
.
Alternatives . . . . . . . . . . . . . . . . . 81
.
6.5 ICMP Port Unreachable Error . . . . . . . . . . . . . 82
.
tcpdump Time Line . . . . . . . . . . . . . . . 84
.

ii

6.6 4.4BSD Processing of ICMP Messages . . . . . . . . . . . 85
.
6.7 Summary . . . . . . . . . . . . . . . . . 87
.
Exercises . . . . . . . . . . . . . . . . . . 87
.
Chapter 7. Ping Program . . . . . . . . . . . . . . . . 88
.
Ping Program . . . . . . . . . . . . . . . . . . 88
.
7.1 Introduction . . . . . . . . . . . . . . . . . 88
.
7.2 Ping Program . . . . . . . . . . . . . . . . 88
.
LAN Output . . . . . . . . . . . . . . . . . 89
.
WAN Output . . . . . . . . . . . . . . . . . 91
.
Hardwired SLIP Links . . . . . . . . . . . . . . . 91
.
Dialup SLIP Links . . . . . . . . . . . . . . . 92
.
7.3 IP Record Route Option . . . . . . . . . . . . . . 93
.
Normal Example . . . . . . . . . . . . . . . . 94
.
Abnormal Output . . . . . . . . . . . . . . . . 96
.
7.4 IP Timestamp Option . . . . . . . . . . . . . . . 97
.
7.5 Summary . . . . . . . . . . . . . . . . . 98
.
Exercises . . . . . . . . . . . . . . . . . . 98
.
Traceroute Program . . . . . . . . . . . . . . . . . 99
.
Traceroute Program . . . . . . . . . . . . . . . . . 99
.
8.1 Introduction . . . . . . . . . . . . . . . . . 99
.
8.2 Traceroute Program Operation . . . . . . . . . . . . . 99
.
8.3 LAN Output . . . . . . . . . . . . . . . . . 100
.
8.4 WAN Output . . . . . . . . . . . . . . . . 104
.
8.5 IP Source Routing Option . . . . . . . . . . . . . . 106
.
traceroute Examples with Loose Source Routing . . . . . . . . 107
.
traceroute Examples with Strict Source Routing . . . . . . . . . 109
.
traceroute Round Trips with Loose Source Routing . . . . . . . 111
.
8.6 Summary . . . . . . . . . . . . . . . . . 112
.
Exercises . . . . . . . . . . . . . . . . . . 112
.
Chapter 9. IP Routing . . . . . . . . . . . . . . . . . 114
.
IP Routing . . . . . . . . . . . . . . . . . . . 114
.
9.1 Introduction . . . . . . . . . . . . . . . . . 114
.
9.2 Routing Principles . . . . . . . . . . . . . . . 114
.
Simple Routing Table . . . . . . . . . . . . . . . 115
.
Initializing a Routing Table . . . . . . . . . . . . . 117
.
A More Complex Routing Table . . . . . . . . . . . . 118
.
No Route to Destination . . . . . . . . . . . . . . 119
.
9.3 ICMP Host and Network Unreachable Errors . . . . . . . . . 119
.
9.4 To Forward or Not to Forward . . . . . . . . . . . . . 120
.
9.5 ICMP Redirect Errors . . . . . . . . . . . . . . 121
.
An Example . . . . . . . . . . . . . . . . . 121
.
More Details . . . . . . . . . . . . . . . . . 123
.
9.6 ICMP Router Discovery Messages . . . . . . . . . . . . 124
.
Router Operation . . . . . . . . . . . . . . . . 125
.
Host Operation . . . . . . . . . . . . . . . . 125
.
Implementation . . . . . . . . . . . . . . . . 126
.
9.7 Summary . . . . . . . . . . . . . . . . . 126
.
Exercises . . . . . . . . . . . . . . . . . . 126
.
Chapter 10. Dynamic Routing Protocols . . . . . . . . . . . . 128
.
Dynamic Routing Protocols . . . . . . . . . . . . . . . 128
.

iii

10.1 Introduction . . . . . . . . . . . . . . . . 128
.
10.2 Dynamic Routing . . . . . . . . . . . . . . . 128
.
10.3 Unix Routing Daemons . . . . . . . . . . . . . . 129
.
10.4 RIP: Routing Information Protocol . . . . . . . . . . . 129
.
Message Format . . . . . . . . . . . . . . . . 129
.
Normal Operation . . . . . . . . . . . . . . . . 130
.
Metrics . . . . . . . . . . . . . . . . . . 131
.
Problems . . . . . . . . . . . . . . . . . . 131
.
Example . . . . . . . . . . . . . . . . . . 132
.
Another Example . . . . . . . . . . . . . . . . 133
.
10.5 RIP Version 2 . . . . . . . . . . . . . . . . 136
.
10.6 OSPF: Open Shortest Path First . . . . . . . . . . . . 136
.
10.7 BGP: Border Gateway Protocol . . . . . . . . . . . . 137
.
10.8 CIDR: Classless Interdomain Routing . . . . . . . . . . . 138
.
10.9 Summary . . . . . . . . . . . . . . . . . 139
.
Exercises . . . . . . . . . . . . . . . . . . 140
.
Chapter 11. UDP: User Datagram Protocol . . . . . . . . . . . . 141
.
UDP: User Datagram Protocol . . . . . . . . . . . . . . 141
.
11.1 Introduction . . . . . . . . . . . . . . . . 141
.
11.2 UDP Header . . . . . . . . . . . . . . . . 141
.
11.3 UDP Checksum . . . . . . . . . . . . . . . . 142
.
tcpdump Output . . . . . . . . . . . . . . . . 143
.
Some Statistics . . . . . . . . . . . . . . . . 144
.
11.4 A Simple Example . . . . . . . . . . . . . . . 144
.
11.5 IP Fragmentation . . . . . . . . . . . . . . . 145
.
11.6 ICMP Unreachable Error (Fragmentation Required) . . . . . . . . 148
.
Example . . . . . . . . . . . . . . . . . . 148
.
11.7 Determining the Path MTU Using Traceroute . . . . . . . . . 150
.
The Worldwide Internet . . . . . . . . . . . . . . 151
.
11.8 Path MTU Discovery with UDP . . . . . . . . . . . . 151
.
Example . . . . . . . . . . . . . . . . . . 151
.
11.9 Interaction Between UDP and ARP . . . . . . . . . . . 155
.
11.10 Maximum UDP Datagram Size . . . . . . . . . . . . 156
.
Datagram Truncation . . . . . . . . . . . . . . . 157
.
11.11 ICMP Source Quench Error . . . . . . . . . . . . . 158
.
11.12 UDP Server Design . . . . . . . . . . . . . . 159
.
Client IP Address and Port Number . . . . . . . . . . . . 160
.
Destination IP Address . . . . . . . . . . . . . . 160
.
UDP Input Queue . . . . . . . . . . . . . . . . 160
.
Restricting Local IP Address . . . . . . . . . . . . . 162
.
Restricting Foreign IP Address . . . . . . . . . . . . . 163
.
Multiple Recipients per Port . . . . . . . . . . . . . 164
.
11.13 Summary . . . . . . . . . . . . . . . . . 165
.
Exercises . . . . . . . . . . . . . . . . . . 165
.
Chapter 12. Broadcasting and Multicasting . . . . . . . . . . . . 167
.
Broadcasting and Multicasting . . . . . . . . . . . . . . 167
.
12.1 Introduction . . . . . . . . . . . . . . . . 167
.
12.2 Broadcasting . . . . . . . . . . . . . . . . 168
.
Limited Broadcast . . . . . . . . . . . . . . . . 168
.
Net-directed Broadcast . . . . . . . . . . . . . . . 169
.

iv

Subnet-directed Broadcast . . . . . . . . . . . . . . 169
.
All-subnets-directed Broadcast . . . . . . . . . . . . . 169
.
12.3 Broadcasting Examples . . . . . . . . . . . . . . 169
.
12.4 Multicasting . . . . . . . . . . . . . . . . 172
.
Multicast Group Addresses . . . . . . . . . . . . . . 172
.
Converting Multicast Group Addresses to Ethernet Addresses . . . . . . 173
.
Multicasting on FDDI and Token Ring Networks . . . . . . . . . 174
.
12.5 Summary . . . . . . . . . . . . . . . . . 175
.
Exercises . . . . . . . . . . . . . . . . . . 175
.
Chapter 13. IGMP: Internet Group Management Protocol . . . . . . . . 176
.
IGMP: Internet Group Management Protocol . . . . . . . . . . . 176
.
13.1 Introduction . . . . . . . . . . . . . . . . 176
.
13.2 IGMP Message . . . . . . . . . . . . . . . . 176
.
13.3 IGMP Protocol . . . . . . . . . . . . . . . . 177
.
Joining a Multicast Group . . . . . . . . . . . . . . 177
.
IGMP Reports and Queries . . . . . . . . . . . . . . 177
.
Implementation Details . . . . . . . . . . . . . . 178
.
Time-to-Live Field . . . . . . . . . . . . . . . 178
.
All-Hosts Group . . . . . . . . . . . . . . . . 179
.
13.4 An Example . . . . . . . . . . . . . . . . 179
.
Multicast Router Example . . . . . . . . . . . . . . 181
.
13.5 Summary . . . . . . . . . . . . . . . . . 183
.
Exercises . . . . . . . . . . . . . . . . . . 183
.
Chapter 14. DNS: The Domain Name System . . . . . . . . . . . 184
.
DNS: The Domain Name System . . . . . . . . . . . . . . 184
.
14.1 Introduction . . . . . . . . . . . . . . . . 184
.
14.2 DNS Basics . . . . . . . . . . . . . . . . 184
.
14.3 DNS Message Format . . . . . . . . . . . . . . 187
.
Question Portion of DNS Query Message . . . . . . . . . . 189
.
Resource Record Portion of DNS Response Message . . . . . . . . 190
.
14.4 A Simple Example . . . . . . . . . . . . . . . 191
.
14.5 Pointer Queries . . . . . . . . . . . . . . . . 195
.
Example . . . . . . . . . . . . . . . . . . 196
.
Hostname Spoofing Check . . . . . . . . . . . . . . 196
.
14.6 Resource Records . . . . . . . . . . . . . . . 197
.
14.7 Caching . . . . . . . . . . . . . . . . . 200
.
14.8 UDP or TCP . . . . . . . . . . . . . . . . 204
.
14.9 Another Example . . . . . . . . . . . . . . . 204
.
14.10 Summary . . . . . . . . . . . . . . . . . 206
.
Exercises . . . . . . . . . . . . . . . . . . 206
.
Chapter 15. TFTP: Trivial File Transfer Protocol . . . . . . . . . . 207
.
TFTP: Trivial File Transfer Protocol . . . . . . . . . . . . . 207
.
15.1 Introduction . . . . . . . . . . . . . . . . 207
.
15.2 Protocol . . . . . . . . . . . . . . . . . 207
.
15.3 An Example . . . . . . . . . . . . . . . . 208
.
15.4 Security . . . . . . . . . . . . . . . . . 210
.
15.5 Summary . . . . . . . . . . . . . . . . . 210
.
Exercises . . . . . . . . . . . . . . . . . . 211
.
Chapter 16. BOOTP: Bootstrap Protocol . . . . . . . . . . . . 212
.
BOOTP: Bootstrap Protocol . . . . . . . . . . . . . . . 212
.

v

16.1 Introduction . . . . . . . . . . . . . . . . 212
.
16.2 BOOTP Packet Format . . . . . . . . . . . . . . 212
.
Port Numbers . . . . . . . . . . . . . . . . . 214
.
16.3 An Example . . . . . . . . . . . . . . . . 214
.
16.4 BOOTP Server Design . . . . . . . . . . . . . . 217
.
16.5 BOOTP Through a Router . . . . . . . . . . . . . 217
.
16.6 Vendor-Specific Information . . . . . . . . . . . . . 218
.
16.7 Summary . . . . . . . . . . . . . . . . . 219
.
Exercises . . . . . . . . . . . . . . . . . . 219
.
Chapter 17. TCP: Transmission Control Protocol . . . . . . . . . . 220
.
TCP: Transmission Control Protocol . . . . . . . . . . . . . 220
.
17.1 Introduction . . . . . . . . . . . . . . . . 220
.
17.2 TCP Services . . . . . . . . . . . . . . . . 220
.
17.3 TCP Header . . . . . . . . . . . . . . . . 221
.
17.4 Summary . . . . . . . . . . . . . . . . . 224
.
Exercises . . . . . . . . . . . . . . . . . . 224
.
Chapter 18. TCP Connection Establishment and Termination . . . . . . . 225
.
TCP Connection Establishment and Termination . . . . . . . . . . 225
.
18.1 Introduction . . . . . . . . . . . . . . . . 225
.
18.2 Connection Establishment and Termination . . . . . . . . . 225
.
tcpdump Output . . . . . . . . . . . . . . . . 225
.
Time Line . . . . . . . . . . . . . . . . . 227
.
Connection Establishment Protocol . . . . . . . . . . . . 227
.
Connection Termination Protocol . . . . . . . . . . . . 228
.
Normal tcpdump Output . . . . . . . . . . . . . . 229
.
18.3 Timeout of Connection Establishment . . . . . . . . . . . 229
.
First Timeout Period . . . . . . . . . . . . . . . 230
.
Type-of-Service Field . . . . . . . . . . . . . . . 231
.
18.4 Maximum Segment Size . . . . . . . . . . . . . . 231
.
18.5 TCP Half-Close . . . . . . . . . . . . . . . . 232
.
18.6 TCP State Transition Diagram . . . . . . . . . . . . 233
.
2MSL Wait State . . . . . . . . . . . . . . . . 234
.
Quiet Time Concept . . . . . . . . . . . . . . . 238
.
FIN WAIT 2 State . . . . . . . . . . . . . . . . 238
.
18.7 Reset Segments . . . . . . . . . . . . . . . . 238
.
Connection Request to Nonexistent Port . . . . . . . . . . . 239
.
Aborting a Connection . . . . . . . . . . . . . . . 239
.
Detecting Half-Open Connections . . . . . . . . . . . . 241
.
18.8 Simultaneous Open . . . . . . . . . . . . . . . 242
.
An Example . . . . . . . . . . . . . . . . . 243
.
18.9 Simultaneous Close . . . . . . . . . . . . . . . 245
.
18.10 TCP Options . . . . . . . . . . . . . . . . 246
.
18.11 TCP Server Design . . . . . . . . . . . . . . 246
.
TCP Server Port Numbers . . . . . . . . . . . . . . 247
.
Restricting Local IP Address . . . . . . . . . . . . . 248
.
Restricting Foreign IP Address . . . . . . . . . . . . . 249
.
Incoming Connection Request Queue . . . . . . . . . . . 250
.
18.12 Summary . . . . . . . . . . . . . . . . . 254
.
Exercises . . . . . . . . . . . . . . . . . . 254
.

vi

Chapter 19. TCP Interactive Data Flow . . . . . . . . . . . . 257
.
TCP Interactive Data Flow . . . . . . . . . . . . . . . 257
.
19.1 Introduction . . . . . . . . . . . . . . . . 257
.
19.2 Interactive Input . . . . . . . . . . . . . . . 257
.
19.3 Delayed Acknowledgments . . . . . . . . . . . . . 259
.
19.4 Nagle Algorithm . . . . . . . . . . . . . . . 260
.
Disabling the Nagle Algorithm . . . . . . . . . . . . . 262
.
An Example . . . . . . . . . . . . . . . . . 262
.
19.5 Window Size Advertisements . . . . . . . . . . . . 267
.
19.6 Summary . . . . . . . . . . . . . . . . . 267
.
Exercises . . . . . . . . . . . . . . . . . . 267
.
Chapter 20. TCP Bulk Data Flow . . . . . . . . . . . . . . 268
.
TCP Bulk Data Flow . . . . . . . . . . . . . . . . . 268
.
20.1 Introduction . . . . . . . . . . . . . . . . 268
.
20.2 Normal Data Flow . . . . . . . . . . . . . . . 268
.
Fast Sender, Slow Receiver . . . . . . . . . . . . . . 271
.
20.3 Sliding Windows . . . . . . . . . . . . . . . 272
.
An Example . . . . . . . . . . . . . . . . . 273
.
20.4 Window Size . . . . . . . . . . . . . . . . 274
.
An Example . . . . . . . . . . . . . . . . . 275
.
20.5 PUSH Flag . . . . . . . . . . . . . . . . . 276
.
Examples . . . . . . . . . . . . . . . . . 277
.
20.6 Slow Start . . . . . . . . . . . . . . . . . 278
.
An Example . . . . . . . . . . . . . . . . . 278
.
20.7 Bulk Data Throughput . . . . . . . . . . . . . . 279
.
Bandwidth-Delay Product . . . . . . . . . . . . . . 282
.
Congestion . . . . . . . . . . . . . . . . . 283
.
20.8 Urgent Mode . . . . . . . . . . . . . . . . 283
.
An Example . . . . . . . . . . . . . . . . . 285
.
20.9 Summary . . . . . . . . . . . . . . . . . 287
.
Exercises . . . . . . . . . . . . . . . . . . 288
.
Chapter 21. TCP Timeout and Retransmission . . . . . . . . . . . 289
.
TCP Timeout and Retransmission . . . . . . . . . . . . . . 289
.
21.1 Introduction . . . . . . . . . . . . . . . . 289
.
21.2 Simple Timeout and Retransmission Example . . . . . . . . . 289
.
21.3 Round-Trip Time Measurement . . . . . . . . . . . . 291
.
Karn’s Algorithm . . . . . . . . . . . . . . . . 293
.
21.4 An RTT Example . . . . . . . . . . . . . . . 293
.
Round-Trip Time Measurements . . . . . . . . . . . . 294
.
RTT Estimator Calculations . . . . . . . . . . . . . 295
.
Slow Start . . . . . . . . . . . . . . . . . 297
.
21.5 Congestion Example . . . . . . . . . . . . . . 297
.
21.6 Congestion Avoidance Algorithm . . . . . . . . . . . . 298
.
21.7 Fast Retransmit and Fast Recovery Algorithms . . . . . . . . . 300
.
21.8 Congestion Example (Continued) . . . . . . . . . . . . 301
.
21.9 Per-Route Metrics . . . . . . . . . . . . . . . 304
.
21.10 ICMP Errors . . . . . . . . . . . . . . . . 304
.
An Example . . . . . . . . . . . . . . . . . 304
.
21.11 Repacketization . . . . . . . . . . . . . . . 308
.
21.12 Summary . . . . . . . . . . . . . . . . . 309
.

vii

Exercises . . . . . . . . . . . . . . . . . . 310
.
Chapter 22. TCP Persist Timer . . . . . . . . . . . . . . 311
.
TCP Persist Timer . . . . . . . . . . . . . . . . . 311
.
22.1 Introduction . . . . . . . . . . . . . . . . 311
.
22.2 An Example . . . . . . . . . . . . . . . . 311
.
22.3 Silly Window Syndrome . . . . . . . . . . . . . . 314
.
An Example . . . . . . . . . . . . . . . . . 314
.
22.4 Summary . . . . . . . . . . . . . . . . . 318
.
Exercises . . . . . . . . . . . . . . . . . . 318
.
Chapter 23. TCP Keepalive Timer . . . . . . . . . . . . . . 319
.
TCP Keepalive Timer . . . . . . . . . . . . . . . . 319
.
23.1 Introduction . . . . . . . . . . . . . . . . 319
.
23.2 Description . . . . . . . . . . . . . . . . . 320
.
23.3 Keepalive Examples . . . . . . . . . . . . . . . 321
.
Other End Crashes . . . . . . . . . . . . . . . 321
.
Other End Crashes and Reboots . . . . . . . . . . . . . 323
.
Other End Is Unreachable . . . . . . . . . . . . . . 324
.
23.4 Summary . . . . . . . . . . . . . . . . . 325
.
Exercises . . . . . . . . . . . . . . . . . . 326
.
Chapter 24. TCP Futures and Performance . . . . . . . . . . . . 327
.
TCP Futures and Performance . . . . . . . . . . . . . . 327
.
24.1 Introduction . . . . . . . . . . . . . . . . 327
.
24.2 Path MTU Discovery . . . . . . . . . . . . . . 327
.
An Example . . . . . . . . . . . . . . . . . 328
.
Big Packets or Small Packets? . . . . . . . . . . . . . 330
.
24.3 Long Fat Pipes . . . . . . . . . . . . . . . . 331
.
Gigabit Networks . . . . . . . . . . . . . . . . 332
.
24.4 Window Scale Option . . . . . . . . . . . . . . 333
.
An Example . . . . . . . . . . . . . . . . . 334
.
24.5 Timestamp Option . . . . . . . . . . . . . . . 336
.
24.6 PAWS: Protection Against Wrapped Sequence Numbers . . . . . . 337
.
24.7 T/TCP: A TCP Extension for Transactions . . . . . . . . . . 338
.
24.8 TCP Performance . . . . . . . . . . . . . . . 341
.
24.9 Summary . . . . . . . . . . . . . . . . . 342
.
Exercises . . . . . . . . . . . . . . . . . . 343
.
Chapter 25. SNMP: Simple Network Management Protocol . . . . . . . . 345
.
SNMP: Simple Network Management Protocol . . . . . . . . . . . 345
.
25.1 Introduction . . . . . . . . . . . . . . . . 345
.
25.2 Protocol . . . . . . . . . . . . . . . . . 345
.
25.3 Structure of Management Information . . . . . . . . . . . 347
.
25.4 Object Identifiers . . . . . . . . . . . . . . . 348
.
25.5 Introduction to the Management Information Base . . . . . . . . 349
.
Case Diagrams . . . . . . . . . . . . . . . . 350
.
25.6 Instance Identification . . . . . . . . . . . . . . 351
.
Simple Variables . . . . . . . . . . . . . . . . 351
.
Tables . . . . . . . . . . . . . . . . . . 351
.
Lexicographic Ordering . . . . . . . . . . . . . . 352
.
25.7 Simple Examples . . . . . . . . . . . . . . . 353
.
Simple Variables . . . . . . . . . . . . . . . . 353
.
get-next Operator . . . . . . . . . . . . . . . 353
.

viii

Table Access . . . . . . . . . . . . . . . . . 354
.
25.8 Management Information Base (Continued) . . . . . . . . . 355
.
system Group . . . . . . . . . . . . . . . . 355
.
interface Group . . . . . . . . . . . . . . . 356
.
at Group . . . . . . . . . . . . . . . . . 358
.
ip Group . . . . . . . . . . . . . . . . . 359
.
icrop Group . . . . . . . . . . . . . . . . 364
.
tcp Group . . . . . . . . . . . . . . . . . 365
.
25.9 Additional Examples . . . . . . . . . . . . . . 368
.
Interface MTU . . . . . . . . . . . . . . . . 368
.
Routing Tables . . . . . . . . . . . . . . . . 369
.
25.10 Traps . . . . . . . . . . . . . . . . . . 370
.
25.11 ASN.1 and BER . . . . . . . . . . . . . . . 372
.
25.12 SNMP Version 2 . . . . . . . . . . . . . . . 373
.
25.13 Summary . . . . . . . . . . . . . . . . . 374
.
Exercises . . . . . . . . . . . . . . . . . . 374
.
Chapter 26. Telnet and Rlogin: Remote Login . . . . . . . . . . . 375
.
Telnet and Rlogin: Remote Login . . . . . . . . . . . . . . 375
.
26.1 Introduction . . . . . . . . . . . . . . . . 375
.
26.2 Rlogin Protocol . . . . . . . . . . . . . . . . 376
.
Application Startup . . . . . . . . . . . . . . . 377
.
Flow Control . . . . . . . . . . . . . . . . . 377
.
Client Interrupt . . . . . . . . . . . . . . . . 378
.
Window Size Changes . . . . . . . . . . . . . . . 378
.
Server to Client Commands . . . . . . . . . . . . . 378
.
Client to Server Commands . . . . . . . . . . . . . 379
.
Client Escapes . . . . . . . . . . . . . . . . 380
.
26.3 Rlogin Examples . . . . . . . . . . . . . . . 380
.
Initial Client-Server Protocol . . . . . . . . . . . . . 380
.
Client Interrupt Key . . . . . . . . . . . . . . . 382
.
26.4 Telnet Protocol . . . . . . . . . . . . . . . . 385
.
NVT ASCII . . . . . . . . . . . . . . . . . 385
.
Telnet Commands . . . . . . . . . . . . . . . . 386
.
Option Negotiation . . . . . . . . . . . . . . . 387
.
Suboption Negotiation . . . . . . . . . . . . . . . 388
.
Half-Duplex, Character at a Time, Line at a Time, or Linemode? . . . . . 389
.
Synch Signal . . . . . . . . . . . . . . . . . 390
.
Client Escapes . . . . . . . . . . . . . . . . 390
.
26.5 Telnet Examples . . . . . . . . . . . . . . . 391
.
Character-at-a-Time Mode . . . . . . . . . . . . . . 391
.
Linemode . . . . . . . . . . . . . . . . . 395
.
Line-at-a-Time Mode (Kludge Line Mode) . . . . . . . . . . 397
.
Linemode: Client Interrupt Key . . . . . . . . . . . . . 400
.
26.6 Summary . . . . . . . . . . . . . . . . . 401
.
Exercises . . . . . . . . . . . . . . . . . . 402
.
Chapter 27. FTP: File Transfer Protocol . . . . . . . . . . . . 403
.
FTP: File Transfer Protocol . . . . . . . . . . . . . . . 403
.
27.1 Introduction . . . . . . . . . . . . . . . . 403
.
27.2 FTP Protocol . . . . . . . . . . . . . . . . 403
.
Data Representation . . . . . . . . . . . . . . . 404
.

ix

FTP Commands . . . . . . . . . . . . . . . . 405
.
FTP Replies . . . . . . . . . . . . . . . . . 406
.
Connection Management . . . . . . . . . . . . . . 408
.
27.3 FTP Examples . . . . . . . . . . . . . . . . 409
.
Connection Management: Ephemeral Data Port . . . . . . . . . 409
.
Connection Management: Default Data Port . . . . . . . . . . 411
.
Text File Transfer: NVT ASCII Representation or Image? . . . . . . . 413
.
Aborting A File Transfer: Telnet Synch Signal . . . . . . . . . 415
.
Anonymous FTP . . . . . . . . . . . . . . . . 417
.
Anonymous FTP from an Unknown IP Address . . . . . . . . . 419
.
27.4 Summary . . . . . . . . . . . . . . . . . 420
.
Exercises . . . . . . . . . . . . . . . . . . 421
.
Chapter 28. SMTP: Simple Mail Transfer Protocol . . . . . . . . . . 422
.
SMTP: Simple Mail Transfer Protocol . . . . . . . . . . . . . 422
.
28.1 Introduction . . . . . . . . . . . . . . . . 422
.
28.2 SMTP Protocol . . . . . . . . . . . . . . . . 422
.
Simple Example . . . . . . . . . . . . . . . . 423
.
SMTP Commands . . . . . . . . . . . . . . . . 426
.
Envelopes, Headers, and Body . . . . . . . . . . . . . 427
.
Relay Agents . . . . . . . . . . . . . . . . . 428
.
NVT ASCII . . . . . . . . . . . . . . . . . 429
.
Retry Intervals . . . . . . . . . . . . . . . . 430
.
28.3 SMTP Examples . . . . . . . . . . . . . . . 430
.
MX Records: Hosts Not Directly Connected to the Internet . . . . . . . 430
.
MX Records: Hosts That Are Down . . . . . . . . . . . . 432
.
VRFY and EXPN Commands . . . . . . . . . . . . . 434
.
28.4 SMTP Futures . . . . . . . . . . . . . . . . 435
.
Envelope Changes: Extended SMTP . . . . . . . . . . . . 435
.
Header Changes: Non-ASCII Characters . . . . . . . . . . . 437
.
Body Changes: Multipurpose Internet Mail Extensions (MIME) . . . . . 439
.
28.5 Summary . . . . . . . . . . . . . . . . . 443
.
Exercises . . . . . . . . . . . . . . . . . . 443
.
Chapter 29. NFS: Network File System . . . . . . . . . . . . . 444
.
NFS: Network File System . . . . . . . . . . . . . . . 444
.
29.1 Introduction . . . . . . . . . . . . . . . . 444
.
29.2 Sun Remote Procedure Call . . . . . . . . . . . . . 444
.
Sun RPC . . . . . . . . . . . . . . . . . . 445
.
29.3 XDR: External Data Representation . . . . . . . . . . . 446
.
29.4 Port Mapper . . . . . . . . . . . . . . . . 446
.
29.5 NFS Protocol . . . . . . . . . . . . . . . . 448
.
File Handles . . . . . . . . . . . . . . . . . 450
.
Mount Protocol . . . . . . . . . . . . . . . . 450
.
NFS Procedures . . . . . . . . . . . . . . . . 451
.
UDP or TCP? . . . . . . . . . . . . . . . . . 452
.
NFS Over TCP . . . . . . . . . . . . . . . . 452
.
29.6 NFS Examples . . . . . . . . . . . . . . . . 453
.
Simple Example: Reading a File . . . . . . . . . . . . 453
.
Simple Example: Creating a Directory . . . . . . . . . . . 455
.
Statelessness . . . . . . . . . . . . . . . . . 456
.
Example: Server Crash . . . . . . . . . . . . . . 456
.

x

Idempotent Procedures . . . . . . . . . . . . . . . 459
.
29.7 NFS Version 3 . . . . . . . . . . . . . . . . 460
.
29.8 Summary . . . . . . . . . . . . . . . . . 460
.
Exercises . . . . . . . . . . . . . . . . . . 461
.
Chapter 30. Other TCP/IP Applications . . . . . . . . . . . . 462
.
Other TCP/IP Applications . . . . . . . . . . . . . . . 462
.
30.1 Introduction . . . . . . . . . . . . . . . . 462
.
30.2 Finger Protocol . . . . . . . . . . . . . . . . 462
.
30.3 Whois Protocol . . . . . . . . . . . . . . . . 464
.
White Pages . . . . . . . . . . . . . . . . . 465
.
30.4 Archie, WAIS, Gopher, Veronica, and WWW . . . . . . . . . 466
.
Archie . . . . . . . . . . . . . . . . . . 466
.
WAIS: Wide Area Information Servers . . . . . . . . . . . 466
.
Gopher . . . . . . . . . . . . . . . . . . 466
.
Veronica: Very Easy Rodent-Oriented Netwide Index to Computerized Archives . . 466
.
WWW: World Wide Web . . . . . . . . . . . . . . 467
.
30.5 X Window System . . . . . . . . . . . . . . . 467
.
Xscope Program . . . . . . . . . . . . . . . . 468
.
LBX: Low Bandwidth X . . . . . . . . . . . . . . 470
.
30.6 Summary . . . . . . . . . . . . . . . . . 470
.
Exercises . . . . . . . . . . . . . . . . . . 470
.
Appendix A: The tcpdump Program . . . . . . . . . . . . . 471
.
The tcpdump Program . . . . . . . . . . . . . . . . 471
.
A.1 BSD Packet Filter . . . . . . . . . . . . . . . 471
.
A.2 SunOS Network Interface Tap . . . . . . . . . . . . 472
.
A.3 SVR4 Data Link Provider Interface . . . . . . . . . . . 473
.
A.4 tcpdump Output . . . . . . . . . . . . . . . 473
.
A.5 Security Considerations . . . . . . . . . . . . . . 474
.
A.6 Socket Debug Option . . . . . . . . . . . . . . 475
.
Appendix B: Computer Clocks . . . . . . . . . . . . . . . 476
.
Computer Clocks . . . . . . . . . . . . . . . . . 476
.
Appendix C: The sock Program . . . . . . . . . . . . . . 478
.
The sock Program . . . . . . . . . . . . . . . . . 478
.
Appendix D: Solutions to Selected Exercises . . . . . . . . . . . 482
.
Solutions to Selected Exercises . . . . . . . . . . . . . . 482
.
Chapter 1 . . . . . . . . . . . . . . . . . . 482
.
Chapter 3 . . . . . . . . . . . . . . . . . . 482
.
Chapter 4 . . . . . . . . . . . . . . . . . . 483
.
Chapter 5 . . . . . . . . . . . . . . . . . . 483
.
Chapter 6 . . . . . . . . . . . . . . . . . . 483
.
Chapter 7 . . . . . . . . . . . . . . . . . . 484
.
Chapter 8 . . . . . . . . . . . . . . . . . . 485
.
Chapter 9 . . . . . . . . . . . . . . . . . . 486
.
Chapter 10 . . . . . . . . . . . . . . . . . . 487
.
Chapter 11 . . . . . . . . . . . . . . . . . . 487
.
Chapter 12 . . . . . . . . . . . . . . . . . . 488
.
Chapter 13 . . . . . . . . . . . . . . . . . . 488
.
Chapter 14 . . . . . . . . . . . . . . . . . . 488
.
Chapter 15 . . . . . . . . . . . . . . . . . . 489
.
Chapter 16 . . . . . . . . . . . . . . . . . . 489
.

xi

Chapter 17 . . . . . . . . . . . . . . . . . . 490
.
Chapter 18 . . . . . . . . . . . . . . . . . . 490
.
Chapter 19 . . . . . . . . . . . . . . . . . . 492
.
Chapter 20 . . . . . . . . . . . . . . . . . . 492
.
Chapter 21 . . . . . . . . . . . . . . . . . . 492
.
Chapter 22 . . . . . . . . . . . . . . . . . . 493
.
Chapter 23 . . . . . . . . . . . . . . . . . . 493
.
Chapter 24 . . . . . . . . . . . . . . . . . . 493
.
Chapter 25 . . . . . . . . . . . . . . . . . . 494
.
Chapter 26 . . . . . . . . . . . . . . . . . . 495
.
Chapter 27 . . . . . . . . . . . . . . . . . . 495
.
Chapter 28 . . . . . . . . . . . . . . . . . . 495
.
Chapter 29 . . . . . . . . . . . . . . . . . . 496
.
Chapter 30 . . . . . . . . . . . . . . . . . . 497
.
Appendix E: Configurable Options . . . . . . . . . . . . . . 498
.
Configurable Options . . . . . . . . . . . . . . . . 498
.
E.1 BSD/386 Version 1.0 . . . . . . . . . . . . . . . 498
.
E.2 SunOS 4.1.3 . . . . . . . . . . . . . . . . 500
.
E.3 System V Release 4 . . . . . . . . . . . . . . . 501
.
E.4 Solaris 2.2 . . . . . . . . . . . . . . . . . 502
.
/dev/ip . . . . . . . . . . . . . . . . . . 503
.
/dev/icmp . . . . . . . . . . . . . . . . . 505
.
/dev/arp . . . . . . . . . . . . . . . . . . 505
.
/dev/udp . . . . . . . . . . . . . . . . . . 505
.
/dev/tcp . . . . . . . . . . . . . . . . . . 506
.
E.5 AIX 3.2.2 . . . . . . . . . . . . . . . . . 508
.
E.6 4.4BSD . . . . . . . . . . . . . . . . . . 509
.
Appendix F: Source Code Availability . . . . . . . . . . . . . 511
.
Source Code Availability . . . . . . . . . . . . . . . . 511
.
RFCs (Section 1.11) . . . . . . . . . . . . . . . 511
.
BSD Net/2 Source Code (Section 1.14) . . . . . . . . . . . 511
.
SLIP (Section 2.4) . . . . . . . . . . . . . . . . 511
.
icmpaddrmask Program (Section 6.3) . . . . . . . . . . . 511
.
icmptime Program (Section 6.4) . . . . . . . . . . . . 511
.
ping Program (Chapter 7) . . . . . . . . . . . . . . 512
.
traceroute Program (Chapter 8) . . . . . . . . . . . . 512
.
Router Discovery Daemon (Section 9.6) . . . . . . . . . . . 512
.
gated Daemon (Section 10.3) . . . . . . . . . . . . . 512
.
traceroute.pmtu Program (Section 11.7) . . . . . . . . . 512
.
IP Multicasting Software (Chapter 13) . . . . . . . . . . . 512
.
BIND Name Server (Chapter 14) . . . . . . . . . . . . 512
.
host Program (Chapter 14) . . . . . . . . . . . . . 512
.
dig and doc Programs (Chapter 14) . . . . . . . . . . . 512
.
BOOTP Server (Chapter 16) . . . . . . . . . . . . . 513
.
TCP High-Speed Extensions (Chapter 24) . . . . . . . . . . 513
.
ISODE SNMP Manager and Agent (Chapter 25) . . . . . . . . . 513
.
MIME Software and Examples (Section 28.4) . . . . . . . . . 513
.
Sun RPC (Section 29.2) . . . . . . . . . . . . . . 513
.
Sun NFS (Chapter 29) . . . . . . . . . . . . . . . 513
.
tcpdump Program (Appendix A) . . . . . . . . . . . . 513
.

xii

BSD Packet Filter (Section A.1) . . . . . . . . . . . . 513
.
sock Program (Appendix C) . . . . . . . . . . . . . 513
.
ttcp Program . . . . . . . . . . . . . . . . 514
.
Author-Written Software . . . . . . . . . . . . . . 514
.
List of Acronyms . . . . . . . . . . . . . . . . . . 515
.
ACRONYMS . . . . . . . . . . . . . . . . . 515
.

xiii

TCP/IP Illustrated,
Volume 1
The Protocols

W. Richard Stevens

Contents
Preface

Chapter 1. Introduction
1.1 Introduction
1.2 Layering
1.3 TCP/IP Layering
1.4 Internet Addresses
1.5 The Domain Name System
1.6 Encapsulation
1.7 Demultiplexing
1.8 Client-Server Model
1.9 Port Numbers
1.10 Standardization Process
1.11 RFCs
1.12 Standard, Simple Services
1.13 The Internet
1.14 Implementations
1.15 Application Programming Interfaces
1.16 Test Network
1.17 Summary

Chapter 2. Link Layer
2.1 Introduction
2.2 Ethernet and IEEE 802 Encapsulation
2.3 Trailer Encapsulation
2.4 SLIP: Serial Line IP
2.5 Compressed SLIP
2.6 PPP: Point-to-Point Protocol
2.7 Loopback Interface
2.8 MTU
2.9 Path MTU
2.10 Serial Line Throughput Calculations
2.11 Summary

Chapter 3. IP: Internet Protocol
3.1 Introduction
3.2 IP Header

1

3.3 IP Routing
3.4 Subnet Addressing
3.5 Subnet Mask
3.6 Special Case IP Address
3.7 A Subnet Example
3.8 ifconfig Command
3.9 netstat Command
3.10 IP Futures
3.11 Summary

Chapter 4. ARP: Address Resolution Protocol
4.1 Introduction
4.2 An Example
4.3 ARP Cache
4.4 ARP Packet Format
4.5 ARP Examples
4.6 Proxy ARP
4.7 Gratuitous ARP
4.8 arp Command
4.9 Summary

Chapter 5. RARP: Reverse Address Resolution Protocol
5.1 Introduction
5.2 RARP Packet Format
5.3 RARP Examples
5.4 RARP Server design
5.5 Summary

Chapter 6. ICMP: Internet Control Message Protocol
6.1 Introduction
6.2 ICMP Message Types
6.3 ICMP Address Mask Request and Reply
6.4 ICMP Timestamp Request and Reply
6.5 ICMP Port Unreachable Error
6.6 4.4BSD Processing of ICMP Messages
6.7 Summary

Chapter 7. Ping Program
7.1 Introduction
7.2 Ping Program
7.3 IP Record Route Option
7.4 IP Timestamp Option
7.5 Summary

Chapter 8. Traceroute Program
8.1 Introduction
8.2 Traceroute Program operation
8.3 LAN output
8.4 WAN output
8.5 IP Source Routing Option

2

8.6 Summary

Chapter 9. IP Routing
9.1 Introduction
9.2 Routing Principles
9.3 ICMP Host and Network Unreachable Errors
9.4 To Forward or Not to Forward
9.5 ICMP Redirect Errors
9.6 ICMP Router Discovery Messages
9.7 Summary

Chapter 10. Dynamic Routing Protocols
10.1 Introduction
10.2 Dynamic Routing
10.3 Unix Routing Daemons
10.4 RIP: Routing Information Protocol
10.5 RIP Version 2
10.6 OSPF: Open Shortest Path First
10.7 BGP: Border Gateway Protocol
10.8 CIDR: Classless Interdomain Routing
10.9 Summary

Chapter 11. UDP: User Datagram Protocol
11.1 Introduction
11.2 UDP Header
11.3 UDP Checksum
11.4 A Simple Example
11.5 IP Fragmentation
11.6 ICMP Unreachable Error (Fragmentation Required)
11.7 Determining the Path MTU Using Traceroute
11.8 Path MTU Discovery with UDP
11.9 Interaction Between UDP and ARP
11.10 Maximum UDP Datagram Size
11.11 ICMP Source Quench Error
11.12 UDP Server Design
11.13 Summary

Chapter 12. Broadcasting and Multicasting
12.1 Introduction
12.2 Broadcasting
12.3 Broadcasting Examples
12.4 Multicasting
12.5 Summary

Chapter 13. IGMP: Internet Group Management Protocol
13.1 Introduction
13.2 IGMP Message
13.3 IGMP Protocol
13.4 An Example
13.5 Summary

3

Chapter 14. DNS: The Domain Name System
14.1 Introduction
14.2 DNS Basics
14.3 DNS Message Format
14.5 Pointer Quiries
14.6 Resourse Records
14.7 Caching
14.8 UDP or TCP
14.9 Another Example
14.10 Summary

Chapter 15. TFTP: Trivial File Transfer Protocol
15.1 Introduction
15.2 Protocol
15.3 An Example
15.4 Security
15.5 Summary

Chapter 16. BOOTP: Bootstrap Protocol
16.1 Introduction
16.2 BOOTP Packet Format
16.3 An Example
16.4 BOOTP Server Design
16.5 BOOTP Through a Router
16.6 Vendor-Specific Information
16.7 Summary

Chapter 17. TCP: Transmission Control Protocol
17.1 Introduction
17.2 TCP Services
17.3 TCP Header
17.4 Summary

Chapter 18. TCP Connection Establishment and Termination
18.1 Introduction
18.2 Connection Establishment and Termination
18.3 Timeout of Connection Establishment
18.4 Maximum Segment Size
18.5 TCP Half-Close
18.6 TCP State Transition Diagram
18.7 Reset Segments
18.8 Simultaneous Open
18.9 Simultaneous Close
18.10 TCP Options
18.11 TCP Server Design
18.12 Summary

4

Chapter 19. TCP Interactive Data Flow
19.1 Introduction
19.2 Interactive Input
19.3 Delayed Acknoledgements
19.4 Nagle Algorithm
19.5 Windows Size Advertisments
19.6 Summary

Chapter 20. TCP Bulk Data Flow
20.1 Introduction
20.2 Normal Data Flow
20.3 Sliding Windows
20.4 Window Size
20.5 PUSH Flag
20.6 Slow Start
20.7 Bulk Data Throughput
20.8 Urgent Mode
20.9 Summary

Chapter 21. TCP Timeout and Retransmission
21.1 Introduction
21.2 Simple Timeout and Retransmission Example
21.3 Round-Trip Time Measurement
21.4 An RTT Example
21.5 Congestion Example
21.6 Congestion Avoidance Algorithm
21.7 Fast Retransmit and Fast Recovery Algorithm
21.8 Congestion Example (Continued)
21.9 Per-Route Metrics
21.10 ICMP Errors
21.11 Repacketization
21.12 Summary

Chapter 22. TCP Persist Timer
22.1 Introduction
22.2 An Example
22.3 Silly Windows Syndrome
22.4 Summary

Chapter 23. TCP Keepalive Timer
23.1 Introduction
23.2 Description
23.3 Keepalive Examples
23.4 Summary

Chapter 24. TCP Futures and Performance
24.1 Introduction
24.2 Path MTU Discovery
24.3 Long Fat Pipes
24.4 Windows Scale Option

5

24.5 Timestamp Option
24.6 PAWS: Protection Against Wrapped Sequence Numbers
24.7 T/TCP: A TCP Extension for Transactions
24.8 TCP Performance
24.9 Summary

Chapter 25. SNMP: Simple Network Management Protocol
25.1 Introduction
25.2 Protocol
25.3 Structure of Management Information
25.4 Object Identifiers
25.5 Introduction to the Management Information Base
25.6 Instance Identification
25.7 Simple Examples
25.8 Management Information Base (Continued)
25.9 Additional Examples
25.10 Traps
25.11 ASN.1 and BER
25.12 SNMP Version 2
25.13 Summary

Chapter 26. Telnet and Rlogin: Remote Login
26.1 Introduction
26.2 Rlogin Protocol
26.3 Rlogin Examples
26.4 Telnet Protocol
26.5 Telnet Examples
26.6 Summary

Chapter 27. FTP: File Transfer Protocol
27.1 Introduction
27.2 FTP Protocol
27.3 FTP Examples
27.4 Summary

Chapter 28. SMTP: Simple Mail Transfer Protocol
28.1 Introduction
28.2 SMTP Protocol
28.3 SMTP Examples
28.4 SMTP Futures
28.5 Summary

Chapter 29. NFS: Network File System
29.1 Introduction
29.2 Sun Remote Procedure Call
29.3 XDR: External Data Representation
29.4 Port Mapper
29.5 NFS Protocol
29.6 NFS Examples
29.7 NFS Version 3

6

29.8 Summary

Chapter 30. Other TCP/IP Applications
30.1 Introduction
30.2 Finger Protocol
30.3 Whois Protocol
30.4 Archie, WAIS, Gopher, Veronica and WWW
30.5 X Window System
30.6 Summary

Appendix A. The tcpdump Program
A.1 BSD Packet Filter
A.2 SunOS Network Interface Tap
A.3 SVR4 Data Link Provider Interface
A.4 tcpdump Output
A.5 Security Considerations
A.6 Socket Debug Option

Appendix B. Computer Clocks

Appendix C. The sock Program

Appendix D. Solutions to Selected Exercises

Appendix E. Configurable Options
E.1 BSD/386 version 1.0
E.2 SunOS 4.1.3
E.3 System V Release 4
E.4 Solaris 2.2
E.5 AIX 3.2.2
E.6 4.4BSD

Appendix F. Source Code Availability

Acronyms

You can write to the designer of this page. All critics are welcome.

7

Preface
Introduction
This book describes the TCP/IP protocol suite, but from a different perspective than other texts on
TCP/IP. Instead of just describing the protocols and what they do, we’ll use a popular diagnostic tool
to watch the protocols in action. Seeing how the protocols operate in varying circumstances provides a
greater understanding of how they work and why certain design decisions were made. It also provides
a look into the implementation of the protocols, without having to wade through thousands of lines of
source code.

When networking protocols were being developed in the 1960s through the 1980s, expensive,
dedicated hardware was required to see the packets going "across the wire." Extreme familiarity with
the protocols was also required to comprehend the packets displayed by the hardware. Functionality of
the hardware analyzers was limited to that built in by the hardware designers.

Today this has changed dramatically with the ability of the ubiquitous workstation to monitor a local
area network [Mogul 1990]. Just attach a workstation to your network, run some publicly available
software (described in Appendix A), and watch what goes by on the wire. While many people consider
this a tool to be used for diagnosing network problems, it is also a powerful tool for understanding
how the network protocols operate, which is the goal of this book.

This book is intended for anyone wishing to understand how the TCP/IP protocols operate:
programmers writing network applications, system administrators responsible for maintaining
computer systems and networks utilizing TCP/IP, and users who deal with TCP/IP applications on a
daily basis.

Organization of the Book
The following figure shows the various protocols and applications that are covered. The italic number
by each box indicates the chapter in which that protocol or application is described.

8

(Numerous fine points are missing from this figure that will be discussed in the appropriate chapter.
For example, both the DNS and RPC use TCP, which we don’t show.)

We take a bottom-up approach to the TCP/IP protocol suite. After providing a basic introduction to
TCP/IP in Chapter 1, we will start at the link layer in Chapter 2 and work our way up the protocol
stack. This provides the required background for later chapters for readers who aren’t familiar with
TCP/IP or networking in general.

This book also uses a functional approach instead of following a strict bottom-to-top order. For
example, Chapter 3 describes the IP layer and the IP header. But there are numerous fields in the IP
header that are best described in the context of an application that uses or is affected by a particular
field. Fragmentation, for example, is best understood in terms of UDP (Chapter 11), the protocol often
affected by it. The time-to-live field is fully described when we look at the Traceroute program in
Chapter 8, because this field is the basis for the operation of the program. Similarly, many features of
ICMP are described in the later chapters, in terms of how a particular ICMP message is used by a
protocol or an application.

We also don’t want to save all the good stuff until the end, so we describe TCP/IP applications as soon
as we have the foundation to understand them. Ping and Trace-route are described after IP and ICMP
have been discussed. The applications built on UDP (multicasting, the DNS, TFTP, and BOOTP) are
described after UDP has been examined. The TCP applications, however, along with network
management, must be saved until the end, after we’ve thoroughly described TCP. This text focuses on
how these applications use the TCP/IP protocols. We do not provide all the details on running these
applications.

Readers
This book is self-contained and assumes no specific knowledge of networking or TCP/IP. Numerous
references are provided for readers interested in additional details on specific topics.

9

This book can be used in many ways. It can be used as a self-study reference and covered from start to
finish by someone interested in all the details on the TCP/IP protocol suite. Readers with some TCP/IP
background might want to skip ahead and start with Chapter 7, and then focus on the specific chapters
in which they’re interested. Exercises are provided at the end of the chapters, and most solutions are in
Appendix D. This is to maximize the usefulness of the text as a self-study reference.

When used as part of a one- or two-semester course in computer networking, the focus should be on IP
(Chapters 3 and 9), UDP (Chapter 11), and TCP (Chapters 17-24), along with some of the application
chapters.

Many forward and backward references are provided throughout the text, along with a thorough index,
to allow individual chapters to be studied by themselves. A list of all the acronyms used throughout
the text, along with the compound term for the acronym, appears on the inside back covers.

If you have access to a network you are encouraged to obtain the software used in this book (Appendix
F) and experiment on your own. Hands-on experimentation with the protocols will provide the greatest
knowledge (and make it more fun).

Systems Used for Testing
Every example in the book was run on an actual network and the resulting output saved in a file for
inclusion in the text. Figure 1.11 shows a diagram of the different hosts, routers, and networks that are
used. (This figure is also duplicated on the inside front cover for easy reference while reading the
book.) This collection of networks is simple enough that the topology doesn’t confuse the examples,
and with four systems acting as routers, we can see the error messages generated by routers.

Most of the systems have a name that indicates the type of software being used: bsdi, svr4, sun,
solaris, aix, slip, and so on. In this way we can identify the type of software that we’re dealing
with by looking at the system name in the printed output.

A wide range of different operating systems and TCP/IP implementations are used:

BSD/386 Version 1.0 from Berkeley Software Design, Inc., on the hosts named bsdi and slip.
This system is derived from the BSD Networking Software, Release 2.0. (We show the lineage of
the various BSD releases in Figure 1.10.)
Unix System V/386 Release 4.0 Version 2.0 from U.H. Corporation, on the host named svr4.
This is vanilla SVR4 and contains the standard implementation of TCP/IP from Lachman
Associates used with most versions of SVR4.
SunOS 4.1.3 from Sun Microsystems, on the host named sun. The SunOS 4.1.x systems are
probably the most widely used TCP/IP implementations. The TCP/IP code is derived from
4.2BSD and 4.3BSD.
Solaris 2.2 from Sun Microsystems, on the host named solaris. The Solaris 2.x systems have a
different implementation of TCP/IP from the earlier SunOS 4.1.x systems, and from SVR4. (This
operating system is really SunOS 5.2, but is commonly called Solaris 2.2.)
AIX 3.2.2 from IBM on the host named aix. The TCP/IP implementation is based on the
4.3BSD Reno release.
4.4BSD from the Computer Systems Research Group at the University of California at Berkeley,
on the host vangogh. cs.berkeley.edu. This system has the latest release of TCP/IP
from Berkeley. (This system isn’t shown in the figure on the inside front cover, but is reachable
across the Internet.)

10

Although these are all Unix systems, TCP/IP is operating system independent, and is available on
almost every popular non-Unix system. Most of this text also applies to these non-Unix
implementations, although some programs (such as Traceroute) may not be provided on all systems.

Typographical Conventions
When we display interactive input and output we’ll show our typed input in a bold font, and the
computer output like this. Comments are added in italics.

bsdi % telnet svr4 discard
connect to the discard server
Trying 140.252.13.34...
this line and next output by Telnet client
Connected to svr4.

Also, we always include the name of the system as part of the shell prompt (bsdi in this example) to
show on which host the command was run.
Throughout the text we’ll use indented, parenthetical notes such as this to describe historical points or implementation
details.

We sometimes refer to the complete description of a command in the Unix manual as in
ifconfig(8). This notation, the name of the command followed by a number in parentheses, is the
normal way of referring to Unix commands. The number in parentheses is the section number in the
Unix manual of the "manual page" for the command, where additional information can be located.
Unfortunately not all Unix systems organize their manuals the same, with regard to the section
numbers used for various groupings of commands. We’ll use the BSD-style section numbers (which is
the same for BSD-derived systems such as SunOS 4.1.3), but your manuals may be organized
differently.

Acknowledgments
Although the author’s name is the only one to appear on the cover, the combined effort of many
people is required to produce a quality text book. First and foremost is the author’s family, who put up
with the long and weird hours that go into writing a book. Thank you once again, Sally, Bill, Ellen,
and David.

The consulting editor, Brian Kernighan, is undoubtedly the best in the business. He was the first one to
read various drafts of the manuscript and mark it up with his infinite supply of red pens. His attention
to detail, his continual prodding for readable prose, and his thorough reviews of the manuscript are an
immense resource to a writer.

Technical reviewers provide a different point of view and keep the author honest by catching technical
mistakes. Their comments, suggestions, and (most importantly) criticisms add greatly to the final
product. My thanks to Steve Bellovin, Jon Crowcroft, Pete Haverlock, and Doug Schmidt for
comments on the entire manuscript. Equally valuable comments were provided on portions of the
manuscript by Dave Borman, Tony DeSimone, Bob Gilligan, Jeff Gitlin, John Gulbenkian, Tom
Herbert, Mukesh Kacker, Barry Margolin, Paul Mockapetris, Burr Nelson, Steve Rago, James Risner,
Chris Walquist, Phil Winterbottom, and Gary Wright. A special thanks to Dave Borman for his
thorough review of all the TCP chapters, and to Bob Gilligan who should be listed as a coauthor for
Appendix E.

11

An author cannot work in isolation, so I would like to thank the following persons for lots of small
favors, especially by answering my numerous e-mail questions: Joe Godsil, Jim Hogue, Mike Karels,
Paul Lucchina, Craig Partridge, Thomas Skibo, and Jerry Toporek.

This book is the result of my being asked lots of questions on TCP/IP for which I could find no quick,
immediate answer. It was then that I realized that the easiest way to obtain the answers was to run
small tests, forcing certain conditions to occur, and just watch what happens. I thank Pete Haverlock
for asking the probing questions and Van Jacobson for providing so much of the publicly available
software that is used in this book to answer the questions.

A book on networking needs a real network to work with along with access to the Internet. My thanks
to the National Optical Astronomy Observatories (NOAO), especially Sidney Wolff, Richard Wolff,
and Steve Grandi, for providing access to their networks and hosts. A special thanks to Steve Grandi
for answering lots of questions and providing accounts on various hosts. My thanks also to Keith
Bostic and Kirk McKu-sick at the U.C. Berkeley CSRG for access to the latest 4.4BSD system.

Finally, it is the publisher that pulls everything together and does whatever is required to deliver the
final product to the readers. This all revolves around the editor, and John Wait is simply the best there
is. Working with John and the rest of the professionals at Addison-Wesley is a pleasure. Their
professionalism and attention to detail show in the end result.

Camera-ready copy of the book was produced by the author, a Troff die-hard, using the Groff package
written by James Clark. I welcome electronic mail from any readers with comments, suggestions, or
bug fixes.

Tucson, Arizona
October 1993

W. Richard Stevens
rstevens@noao.edu
http://www.noao.edu/~rstevens

12

Introduction
1.1 Introduction
The TCP/IP protocol suite allows computers of all sizes, from many different computer vendors,
running totally different operating systems, to communicate with each other. It is quite amazing
because its use has far exceeded its original estimates. What started in the late 1960s as a
government-financed research project into packet switching networks has, in the 1990s, turned into the
most widely used form of networking between computerrs. It is truly an open system in that the
definition of the protocol suite and many of its implementations are publicly available at little or no
charge. It forms the basis for what is called the worldwide Internet, or the Internet, a wide area
network (WAN) of more than one million computers that literally spans the globe.

This chapter provides an overview of the TCP/IP protocol suite, to establish an adequate background
for the remaining chapters. For a historical perspective on the early development of TCP/IP see [Lynch
1993].

1.2 Layering
Networking protocols are normally developed in layers, with each layer responsible for a different
facet of the communications. A protocol suite, such as TCP/IP, is the combination of different
protocols at various layers. TCP/IP is normally considered to be a 4-layer system, as shown in Figure
1.1.

Application Telnet, FTP, e-mail, etc.
Transport TCP, UDP
Network IP, ICMP, IGMP
Link device driver and interface card

Figure 1.1 The four layers of the TCP/IP protocol suite.

Each layer has a different responsibility.

1. The link layer, sometimes called the data-link layer or network interface layer, normally includes
the device driver in the operating system and the corresponding network interface card in the
computer. Together they handle all the hardware details of physically interfacing with the cable
(or whatever type of media is being used).
2. The network layer (sometimes called the internet layer) handles the movement of packets around
the network. Routing of packets, for example, takes place here. IP (Internet Protocol), ICMP
(Internet Control Message Protocol), and IGMP (Internet Group Management Protocol) provide
the network layer in the TCP/IP protocol suite.
3. The transport layer provides a flow of data between two hosts, for the application layer above. In
the TCP/IP protocol suite there are two vastly different transport protocols: TCP (Transmission
Control Protocol) and UDP (User Datagram Protocol).

TCP provides a reliable flow of data between two hosts. It is concerned with things such as
dividing the data passed to it from the application into appropriately sized chunks for the network
layer below, acknowledging received packets, setting timeouts to make certain the other end

13

acknowledges packets that are sent, and so on. Because this reliable flow of data is provided by
the transport layer, the application layer can ignore all these details.

UDP, on the other hand, provides a much simpler service to the application layer. It just sends
packets of data called datagrams from one host to the other, but there is no guarantee that the
datagrams reach the other end. Any desired reliability must be added by the application layer.

There is a use for each type of transport protocol, which we’ll see when we look at the different
applications that use TCP and UDP.
4. The application layer handles the details of the particular application. There are many common
TCP/IP applications that almost every implementation provides:
Telnet for remote login,
FTP, the File Transfer Protocol,
SMTP, the Simple Mail Transfer protocol, for electronic mail,
SNMP, the Simple Network Management Protocol,

and many more, some of which we cover in later chapters.

If we have two hosts on a local area network (LAN) such as an Ethernet, both running FTP, Figure 1.2
shows the protocols involved.

Figure 1.2 Two hosts on a LAN running FTP.

We have labeled one application box the FTP client and the other the FTP server. Most network
applications are designed so that one end is the client and the other side the server. The server provides
some type of service to clients, in this case access to files on the server host. In the remote login
application, Telnet, the service provided to the client is the ability to login to the server’s host.

Each layer has one or more protocols for communicating with its peer at the same layer. One protocol,
for example, allows the two TCP layers to communicate, and another protocol lets the two IP layers
communicate.

14

On the right side of Figure 1.2 we have noted that normally the application layer is a user process
while the lower three layers are usually implemented in the kernel (the operating system). Although
this isn’t a requirement, it’s typical and this is the way it’s done under Unix.

There is another critical difference between the top layer in Figure 1.2 and the lower three layers. The
application layer is concerned with the details of the application and not with the movement of data
across the network. The lower three layers know nothing about the application but handle all the
communication details.

We show four protocols in Figure 1.2, each at a different layer. FTP is an application layer protocol,
TCP is a transport layer protocol, IP is a network layer protocol, and the Ethernet protocols operate at
the link layer. The TCP/IP protocol suite is a combination of many protocols. Although the commonly
used name for the entire protocol suite is TCP/IP, TCP and IP are only two of the protocols. (An
alternative name is the Internet Protocol Suite.)

The purpose of the network interface layer and the application layer are obvious-the former handles
the details of the communication media (Ethernet, token ring, etc.) while the latter handles one specific
user application (FTP, Telnet, etc.). But on first glance the difference between the network layer and
the transport layer is somewhat hazy. Why is there a distinction between the two? To understand the
reason, we have to expand our perspective from a single network to a collection of networks.

One of the reasons for the phenomenal growth in networking during the 1980s was the realization that
an island consisting of a stand-alone computer made little sense. A few stand-alone systems were
collected together into a network. While this was progress, during the 1990s we have come to realize
that this new, bigger island consisting of a single network doesn’t make sense either. People are
combining multiple networks together into an internetwork, or an internet. An internet is a collection
of networks that all use the same protocol suite.

The easiest way to build an internet is to connect two or more networks with a router. This is often a
special-purpose hardware box for connecting networks. The nice thing about routers is that they
provide connections to many different types of physical networks: Ethernet, token ring, point-to-point
links, FDDI (Fiber Distributed Data Interface), and so on.

These boxes are also called IP routers, but we’ll use the term router.

Historically these boxes were called gateways, and this term is used throughout much of the TCP/IP literature. Today the
term gateway is used for an application gateway: a process that connects two different protocol suites (say, TCP/IP and
IBM’s SNA) for one particular application (often electronic mail or file transfer).

Figure 1.3 shows an internet consisting of two networks: an Ethernet and a token ring, connected with
a router. Although we show only two hosts communicating, with the router connecting the two
networks, any host on the Ethernet can communicate with any host on the token ring.

In Figure 1.3 we can differentiate between an end system (the two hosts on either side) and an
intermediate system (the router in the middle). The application layer and the transport layer use
end-to-end protocols. In our picture these two layers are needed only on the end systems. The network
layer, however, provides a hop-by-hop protocol and is used on the two end systems and every
intermediate system.

15

Figure 1.3 Two networks connected with a router.

In the TCP/IP protocol suite the network layer, IP, provides an unreliable service. That is, it does its
best job of moving a packet from its source to its final destination, but there are no guarantees. TCP,
on the other hand, provides a reliable transport layer using the unreliable service of IP To provide this
service, TCP performs timeout and retransmission, sends and receives end-to-end acknowledgments,
and so on. The transport layer and the network layer have distinct responsibilities.

A router, by definition, has two or more network interface layers (since it connects two or more
networks). Any system with multiple interfaces is called multihomed. A host can also be multihomed
but unless it specifically forwards packets from one interface to another, it is not called a router. Also,
routers need not be special hardware boxes that only move packets around an internet. Most TCP/IP
implementations allow a multihomed host to act as a router also, but the host needs to be specifically
configured for this to happen. In this case we can call the system either a host (when an application
such as FTP or Telnet is being used) or a router (when it’s forwarding packets from one network to
another). We’ll use whichever term makes sense given the context.

One of the goals of an internet is to hide all the details of the physical layout of the internet from the
applications. Although this isn’t obvious from our two-network internet in Figure 1.3, the application
layers can’t care (and don’t care) that one host is on an Ethernet, the other on a token ring, with a
router between. There could be 20 routers between, with additional types of physical interconnections,
and the applications would run the same. This hiding of the details is what makes the concept of an
internet so powerful and useful.

Another way to connect networks is with a bridge. These connect networks at the link layer, while
routers connect networks at the network layer. Bridges makes multiple LANs appear to the upper
layers as a single LAN.

16

TCP/IP internets tend to be built using routers instead of bridges, so we’ll focus on routers. Chapter 12
of [Perlman 1992] compares routers and bridges.

1.3 TCP/IP Layering
There are more protocols in the TCP/IP protocol suite. Figure 1.4 shows some of the additional
protocols that we talk about in this text.

Figure 1.4 Various protocols at the different layers in the TCP/IP protocol suite.

TCP and UDP are the two predominant transport layer protocols. Both use IP as the network layer.

TCP provides a reliable transport layer, even though the service it uses (IP) is unreliable. Chapters 17
through 22 provide a detailed look at the operation of TCP. We then look at some TCP applications:
Telnet and Riogin in Chapter 26, FTP in Chapter 27, and SMTP in Chapter 28. The applications are
normally user processes.

UDP sends and receives datagrams for applications. A datagram is a unit of information (i.e., a certain
number of bytes of information that is specified by the sender) that travels from the sender to the
receiver. Unlike TCP, however, UDP is unreliable. There is no guarantee that the datagram ever gets
to its final destination. Chapter 11 looks at UDP, and then Chapter 14 (the Domain Name System),
Chapter 15 (the Trivial File Transfer Protocol), and Chapter 16 (the Bootstrap Protocol) look at some
applications that use UDP. SNMP (the Simple Network Management Protocol) also uses UDP, but
since it deals with many of the other protocols, we save a discussion of it until Chapter 25.

17

IP is the main protocol at the network layer. It is used by both TCP and UDP. Every piece of TCP and
UDP data that gets transferred around an internet goes through the IP layer at both end systems and at
every intermediate router. In Figure 1.4 we also show an application accessing IP directly. This is rare,
but possible. (Some older routing protocols were implemented this way. Also, it is possible to
experiment with new transport layer protocols using this feature.) Chapter 3 looks at IP, but we save
some of the details for later chapters where their discussion makes more sense. Chapters 9 and 10 look
at how IP performs routing.

ICMP is an adjunct to IP. It is used by the IP layer to exchange error messages and other vital
information with the IP layer in another host or router. Chapter 6 looks at ICMP in more detail.
Although ICMP is used primarily by IP, it is possible for an application to also access it. Indeed we’ll
see that two popular diagnostic tools, Ping and Traceroute (Chapters 7 and 8), both use ICMP.

IGMP is the Internet Group Management Protocol. It is used with multicasting: sending a UDP
datagram to multiple hosts. We describe the general properties of broadcasting (sending a UDP
datagram to every host on a specified network) and multicasting in Chapter 12, and then describe
IGMP itself in Chapter 13.

ARP (Address Resolution Protocol) and RARP (Reverse Address Resolution Protocol) are specialized
protocols used only with certain types of network interfaces (such as Ethernet and token ring) to
convert between the addresses used by the IP layer and the addresses used by the network interface.
We examine these protocols in Chapters 4 and 5, respectively.

1.4 Internet Addresses
Every interface on an internet must have a unique Internet address (also called an IP address). These
addresses are 32-bit numbers. Instead of using a flat address space such as 1, 2, 3, and so on, there is a
structure to Internet addresses. Figure 1.5 shows the five different classes of Internet addresses.

These 32-bit addresses are normally written as four decimal numbers, one for each byte of the address.
This is called dotted-decimal notation. For example, the class B address of the author’s primary system
is 140.252.13.33.

The easiest way to differentiate between the different classes of addresses is to look at the first number
of a dotted-decimal address. Figure 1.6 shows the different classes, with the first number in boldface.

18

Figure 1.5 The five different classes of Internet addresses.

Class Range
A 0.0.0.0 to 127.255.255.255
B 128.0.0.0 to 191.255.255.255
C 192.0.0.0 to 223.255.255.255
D 224.0.0.0 to 239.255.255.255
E 240.0.0.0 to 247.255.255.255

Figure 1.6 Ranges for different classes of IP addresses.

It is worth reiterating that a multihomed host will have multiple IP addresses: one per interface.

Since every interface on an internet must have a unique IP address, there must be one central authority
for allocating these addresses for networks connected to the worldwide Internet. That authority is the
Internet Network Information Center, called the InterNIC. The InterNIC assigns only network IDs.
The assignment of host IDs is up to the system administrator.

Registration services for the Internet (IP addresses and DNS domain names) used to be handled by the NIC, at
nic.ddn.mil. On April 1, 1993, the InterNIC was created. Now the NIC handles these requests only for the Defense
Data Network (DDN). All other Internet users now use the InterNIC registration services, at rs.internic.net.

There are actually three parts to the InterNIC: registration services (rs.internic.net), directory and database services
(ds.internic.net), and information services (is.internic.net). See Exercise 1.8 for additional information on
the InterNIC.

There are three types of IP addresses: unicast (destined for a single host), broadcast (destined for all
hosts on a given network), and multicast (destined for a set of hosts that belong to a multicast group).
Chapters 12 and 13 look at broadcasting and multicasting in more detail.

In Section 3.4 we’ll extend our description of IP addresses to include subnetting, after describing IP
routing. Figure 3.9 shows the special case IP addresses: host IDs and network IDs of all zero bits or all
one bits.

1.5 The Domain Name System
Although the network interfaces on a host, and therefore the host itself, are known by IP addresses,
humans work best using the name of a host. In the TCP/IP world the Domain Name System (DNS) is a
distributed database that provides the mapping between IP addresses and hostnames. Chapter 14 looks
into the DNS in detail.

For now we must be aware that any application can call a standard library function to look up the IP
address (or addresses) corresponding to a given hostname. Similarly a function is provided to do the
reverse lookup-given an IP address, look up the corresponding hostname.

Most applications that take a hostname as an argument also take an IP address. When we use the
Telnet client in Chapter 4, for example, one time we specify a host-name and another time we specify
an IP address.

19

1.6 Encapsulation
When an application sends data using TCP, the data is sent down the protocol stack, through each
layer, until it is sent as a stream of bits across the network. Each layer adds information to the data by
prepending headers (and sometimes adding trailer information) to the data that it receives. Figure 1.7
shows this process. The unit of data that TCP sends to IP is called a TCP segment. The unit of data that
IP sends to the network interface is called an IP datagram. The stream of bits that flows across the
Ethernet is called a frame.

The numbers at the bottom of the headers and trailer of the Ethernet frame in Figure 1.7 are the typical
sizes of the headers in bytes. We’ll have more to say about each of these headers in later sections.

A physical property of an Ethernet frame is that the size of its data must be between 46 and 1500
bytes. We’ll encounter this minimum in Section 4.5 and we cover the maximum in Section 2.8.

All the Internet standards and most books on TCP/IP use the term octet instead of byte. The use of this cute, but baroque
term is historical, since much of the early work on TCP/IP was done on systems such as the DEC-10, which did not use
8-bit bytes. Since almost every current computer system uses 8-bit bytes, we’ll use the term byte in this text.

To be completely accurate in Figure 1.7 we should say that the unit of data passed between IP and the network interface is
a packet. This packet can be either an IP datagram or a fragment of an IP datagram. We discuss fragmentation in detail in
Section 11.5.

We could draw a nearly identical picture for UDP data. The only changes are that the unit of
information that UDP passes to IP is called a UDP datagram, and the size of the UDP header is 8
bytes.

20

Figure 1.7 Encapsulation of data as it goes down the protocol stack.

Recall from Figure 1.4 that TCP, UDP, ICMP, and IGMP all send data to IP. IP must add some type of
identifier to the IP header that it generates, to indicate the layer to which the data belongs. IP handles
this by storing an 8-bit value in its header called the protocol field. A value of 1 is for ICMP, 2 is for
IGMP, 6 indicates TCP, and 17 is for UDP.

Similarly, many different applications can be using TCP or UDP at any one time. The transport layer
protocols store an identifier in the headers they generate to identify the application. Both TCP and
UDP use 16-bit port numbers to identify applications. TCP and UDP store the source port number and
the destination port number in their respective headers.

The network interface sends and receives frames on behalf of IP, ARP, and RARP. There must be
some form of identification in the Ethernet header indicating which network layer protocol generated
the data. To handle this there is a 16-bit frame type field in the Ethernet header.

1.7 Demultiplexing
When an Ethernet frame is received at the destination host it starts its way up the protocol stack and all
the headers are removed by the appropriate protocol box. Each protocol box looks at certain identifiers
in its header to determine which box in the next upper layer receives the data. This is called
demultiplexing. Figure 1.8 shows how this takes place.

Figure 1.8 The demultiplexing of a received Ethernet frame.

Positioning the protocol boxes labeled "ICMP" and "IGMP" is always a challenge. In Figure 1.4 we showed them at the
same layer as IP, because they really are adjuncts to IP. But here we show them above IP, to reiterate that ICMP messages
and IGMP messages are encapsulated in IP datagrams.

21

We have a similar problem with the boxes "ARP" and "RARP." Here we show them above the Ethernet device driver
because they both have their own Ethernet frame types, like IP datagrams. But in Figure 2.4 we’ll show ARP as part of the
Ethernet device driver, beneath IP, because that’s where it logically fits.

Realize that these pictures of layered protocol boxes are not perfect.

When we describe TCP in detail we’ll see that it really demultiplexes incoming segments using the
destination port number, the source IP address, and the source port number.

1.8 Client-Server Model
Most networking applications are written assuming one side is the client and the other the server. The
purpose of the application is for the server to provide some defined service for clients.

We can categorize servers into two classes: iterative or concurrent. An iterative server iterates through
the following steps.

I1. Wait for a client request to arrive.

I2. Process the client request.

I3. Send the response back to the client that sent the request.

I4. Go back to step I1.

The problem with an iterative server is when step I2 takes a while. During this time no other clients are
serviced. A concurrent server, on the other hand, performs the following steps.

Cl. Wait for a client request to arrive.

C2. Start a new server to handle this client’s request. This may involve creating a new process, task, or
thread, depending on what the underlying operating system supports. How this step is performed
depends on the operating system.

This new server handles this client’s entire request. When complete, this new server terminates.

C3. Go back to step Cl.

The advantage of a concurrent server is that the server just spawns other servers to handle the client
requests. Each client has, in essence, its own server. Assuming the operating system allows
multiprogramming, multiple clients are serviced concurrently.

The reason we categorize servers, and not clients, is because a client normally can’t tell whether it’s
talking to an iterative server or a concurrent server.

As a general rule, TCP servers are concurrent, and UDP servers are iterative, but there are a few
exceptions. We’ll look in detail at the impact of UDP on its servers in Section 11.12, and the impact of
TCP on its servers in Section 18.11.

22

1.9 Port Numbers
We said that TCP and UDP identify applications using 16-bit port numbers. How are these port
numbers chosen?

Servers are normally known by their well-known port number. For example, every TCP/IP
implementation that provides an FTP server provides that service on TCP port 21. Every Telnet server
is on TCP port 23. Every implementation of TFTP (the Trivial File Transfer Protocol) is on UDP port
69. Those services that can be provided by any implementation of TCP/IP have well-known port
numbers between 1 and 1023. The well-known ports are managed by the Internet Assigned Numbers
Authority (IANA).

Until 1992 the well-known ports were between I and 255. Ports between 256 and 1023 were normally used by Unix
systems for Unix-specific services-that is, services found on a Unix system, but probably not found on other operating
systems. The IANA now manages the ports between 1 and 1023.

An example of the difference between an Internet-wide service and a Unix-specific service is the difference between Telnet
and Riogin. Both allow us to login across a network to another host. Telnet is a TCP/IP standard with a well-known port
number of 23 and can be implemented on almost any operating system. Rlogin, on the other hand, was originally designed
for Unix systems (although many non-Unix systems now provide it also) so its well-known port was chosen in the early
1980s as 513.

A client usually doesn’t care what port number it uses on its end. All it needs to be certain of is that
whatever port number it uses be unique on its host. Client port numbers are called ephemeral ports
(i.e., short lived). This is because a client typically exists only as long as the user running the client
needs its service, while servers typically run as long as the host is up.

Most TCP/IP implementations allocate ephemeral port numbers between 1024 and 5000. The port
numbers above 5000 are intended for other servers (those that aren’t well known across the Internet).
We’ll see many examples of how ephemeral ports are allocated in the examples throughout the text.

Solaris 2.2 is a notable exception. By default the ephemeral ports for TCP and UDP start at 32768. Section E.4 details the
configuration options that can be modified by the system administrator to change these defaults.

The well-known port numbers are contained in the file /etc/services on most Unix systems. To find
the port numbers for the Telnet server and the Domain Name System, we can execute

sun % grep telnet /etc/services
says it uses TCP port 23
telnet 23/tcp
sun % grep domain /etc/services
domain 53/udp says it uses UDP port 53
domain 53/tcp and TCP port 53

Reserved Ports
Unix systems have the concept of reserved ports. Only a process with superuser privileges can assign
itself a reserved port.

These port numbers are in the range of 1 to 1023, and are used by some applications (notably Rlogin,
Section 26.2), as part of the authentication between the client and server.

23

1.10 Standardization Process
Who controls the TCP/IP protocol suite, approves new standards, and the like? There are four groups
responsible for Internet technology.

1. The Internet Society (ISOC) is a professional society to facilitate, support, and promote the
evolution and growth of the Internet as a global research communications infrastructure.
2. The Internet Architecture Board (IAB) is the technical oversight and coordination body. It is
composed of about 15 international volunteers from various disciplines and serves as the final
editorial and technical review board for the quality of Internet standards. The IAB falls under the
ISOC.
3. The Internet Engineering Task Force (IETF) is the near-term, standards-oriented group, divided
into nine areas (applications, routing and addressing, security, etc.). The IETF develops the
specifications that become Internet standards. An additional Internet Engineering Steering Group
(IESG) was formed to help the IETF chair.
4. The Internet Research Task Force (IRTF) pursues long-term research projects.

Both the IRTF and the IETF fall under the IAB. [Crocker 1993] provides additional details on the
standardization process within the Internet, as well as some of its early history.

1.11 RFCs
All the official standards in the internet community are published as a Request for Comment, or RFC.
Additionally there are lots of RFCs that are not official standards, but are published for informational
purposes. The RFCs range in size from I page to almost 200 pages. Each is identified by a number,
such as RFC 1122, with higher numbers for newer RFCs.

All the RFCs are available at no charge through electronic mail or using FTP across the Internet.
Sending electronic mail as shown here:

To: rfc-info@OISI.EDU
Subject: getting rfcs

help: ways_to_get_rfcs

returns a detailed listing of various ways to obtain the RFCs.

The latest RFC index is always a starting point when looking for something. This index specifies when
a certain RFC has been replaced by a newer RFC, and if a newer RFC updates some of the information
in that RFC. There are a few important RFCs.

1. The Assigned Numbers RFC specifies all the magic numbers and constants that are used in the
Internet protocols. At the time of this writing the latest version of this RFC is 1340 [Reynolds and
Postel 1992]. All the Internet-wide well-known ports are listed here.

When this RFC is updated (it is normally updated at least yearly) the index listing for 1340 will
indicate which RFC has replaced it.
2. The Internet Official Protocol Standards, currently RFC 1600 [Postel 1994]. This RFC specifies
the state of standardization of the various Internet protocols. Each protocol has one of the
following states of standardization: standard, draft standard, proposed standard, experimental,
informational, or historic. Additionally each protocol has a requirement level: required,

24

recommended, elective, limited use, or not recommended.

Like the Assigned Numbers RFC, this RFC is also reissued regularly. Be sure you’re reading the
current copy.
3. The Host Requirements RFCs, 1122 and 1123 [Braden 1989a, 1989b]. RFC 1122 handles the link
layer, network layer, and transport layer, while RFC 1123 handles the application layer. These two
RFCs make numerous corrections and interpretations of the important earlier RFCs, and are often the
starting point when looking at any of the finer details of a given protocol. They list the features and
implementation details of the protocols as either "must," "should," "may," "should not," or "must not."

[Borman 1993b] provides a practical look at these two RFCs, and RFC 1127 [Braden 1989c]
provides an informal summary of the discussions and conclusions of the working group that developed
the Host Requirements RFCs.
4. The Router Requirements RFC. The official version of this is RFC 1009 [Braden and Postel
1987], but a new version is nearing completion [Almquist 1993]. This is similar to the host
requirements RFCs, but specifies the unique requirements of routers.

1.12 Standard, Simple Services
There are a few standard, simple services that almost every implementation provides. We’ll use some
of these servers throughout the text, usually with the Telnet client. Figure 1.9 describes these services.
We can see from this figure that when the same service is provided using both TCP and UDP, both
port numbers are normally chosen to be the same.

If we examine the port numbers for these standard services and other standard TCP/IP services (Telnet, FTP, SMTP, etc.),
most are odd numbers. This is historical as these port numbers are derived from the NCP port numbers. (NCP, the Network
Control Protocol, preceded TCP as a transport layer protocol for the ARPANET.) NCP was simplex, not full-duplex, so
each application required two connections, and an even-odd pair of port numbers was reserved for each application. When
TCP and UDP became the standard transport layers, only a single port number was needed per application, so the odd port
numbers from NCP were used.

TCP UDP
Name RFC Description
port port
echo 7 7 862 Server returns whatever the client sends.
discard 9 9 863 Server discards whatever the client sends.
daytime 13 13 867 Server returns the time and date in a human-readable format.
TCP server sends a continual stream of characters, until the
connection is terminated by the client. UDP server sends a
chargen 19 19 864
datagram containing a random number of characters each time the
client sends a datagram.
Server returns the time as a 32-bit binary number. This number
time 37 37 868 represents the number of seconds since midnight January 1, 1900,
UTC.

Figure 1.9 Standard, simple services provided by most implementations.

25

1.13 The Internet
In Figure 1.3 we showed an internet composed of two networks - an Ethernet and a token ring. In
Sections 1.4 and 1.9 we talked about the worldwide Internet and the need to allocate IP addresses
centrally (the InterNIC) and the well-known port numbers (the IANA). The word internet means
different things depending on whether it’s capitalized or not.

The lowercase internet means multiple networks connected together, using a common protocol suite.
The uppercase Internet refers to the collection of hosts (over one million) around the world that can
communicate with each other using TCP/IP. While the Internet is an internet, the reverse is not true.

1.14 Implementations
The de facto standard for TCP/IP implementations is the one from the Computer Systems Research
Group at the University of California at Berkeley. Historically this has been distributed with the 4.x
BSD system (Berkeley Software Distribution), and with the "BSD Networking Releases." This source
code has been the starting point for many other implementations.

Figure 1.10 shows a chronology of the various BSD releases, indicating the important TCP/IP features.
The BSD Networking Releases shown on the left side are publicly available source code releases
containing all of the networking code: both the protocols themselves and many of the applications and
utilities (such as Telnet and FTP).

Throughout the text we’ll use the term Berkeley-derived implementation to refer to vendor
implementations such as SunOS 4.x, SVR4, and AIX 3.2 that were originally developed from the
Berkeley sources. These implementations have much in common, often including the same bugs!

26

Figure 1.10 Various BSD releases with important TCP/IP features.

Much of the original research in the Internet is still being applied to the Berkeley system-new
congestion control algorithms (Section 21.7), multicasting (Section 12.4), "long fat pipe"
modifications (Section 24.3), and the like.

1.15 Application Programming Interfaces
Two popular application programming interfaces (APIs) for applications using the TCP/IP protocols
are called sockets and TLI (Transport Layer Interface). The former is sometimes called "Berkeley
sockets," indicating where it was originally developed. The latter, originally developed by AT&T, is
sometimes called XTI (X/Open Transport Interface), recognizing the work done by X/Open, an
international group of computer vendors that produce their own set of standards. XTI is effectively a
superset of TLI.

This text is not a programming text, but occasional reference is made to features of TCP/IP that we
look at, and whether that feature is provided by the most popular API (sockets) or not. All the
programming details for both sockets and TLI are available in [Stevens 1990].

27

1.16 Test Network
Figure 1.11 shows the test network that is used for all the examples in the text. This figure is also
duplicated on the inside front cover for easy reference while reading the book.

Figure 1.11 Test network used for all the examples in the text. All IP addresses begin with 140.252.

Most of the examples are run on the lower four systems in this figure (the author’s subnet). All the IP
addresses in this figure belong to the class B network ID 140.252. All the hostnames belong to the
.tuc.noao.edu domain.(noao stands for "National Optical Astronomy Observatories" and tuc
stands for Tucson.) For example, the lower right system has a complete hostname of
svr4.tuc.noao.edu and an IP address of 140.252.13.34. The notation at the top of each box is
the operating system running on that system. This collection of systems and networks provides hosts
and routers running a variety of TCP/IP implementations.

It should be noted that there are many more networks and hosts in the noao.edu domain than we
show in Figure 1.11. All we show here are the systems that we’ll encounter throughout the text.

In Section 3.4 we describe the form of subnetting used on this network, and in Section 4.6 we’ll
provide more details on the dial-up SLIP connection between sun and netb. Section 2.4 describes
SLIP in detail.

1.17 Summary
This chapter has been a whirlwind tour of the TCP/IP protocol suite, introducing many of the terms
and protocols that we discuss in detail in later chapters.

28

The four layers in the TCP/IP protocol suite are the link layer, network layer, transport layer, and
application layer, and we mentioned the different responsibilities of each. In TCP/IP the distinction
between the network layer and the transport layer is critical: the network layer (IP) provides a
hop-by-hop service while the transport layers (TCP and UDP) provide an end-to-end service.

An internet is a collection of networks. The common building block for an internet is a router that
connects the networks at the IP layer. The capital-l Internet is an internet that spans the globe and
consists of more than 10,000 networks and more than one million computers.

On an internet each interface is identified by a unique IP address, although users tend to use hostnames
instead of IP addresses. The Domain Name System provides a dynamic mapping between hostnames
and IP addresses. Port numbers are used to identify the applications communicating with each other
and we said that servers use well-known ports while clients use ephemeral ports.

Exercises
1.1 Calculate the maximum number of class A, B, and C network IDs.

1.2 Fetch the file nsfnet/statistics/history.netcount using anonymous FTP (Section
27.3) from the host nic.merit.edu. This file contains the number of domestic and foreign
networks announced to the NSFNET infrastructure. Plot these values with the year on the x-axis and a
logarithmic y-axis with the total number of networks. The maximum value for the y-axis should be the
value calculated in the previous exercise. If the data shows a visual trend, extrapolate the values to
estimate when the current addressing scheme will run out of network IDs. (Section 3.10 talks about
proposals to correct this problem.)

1.3 Obtain a copy of the Host Requirements RFC [Braden 1989a] and look up the robustness principle
that applies to every layer of the TCP/IP protocol suite. What is the reference for this principle?

1.4 Obtain a copy of the latest Assigned Numbers RFC. What is the well-known port for the "quote of
the day" protocol? Which RFC defines the protocol?

1.5 If you have an account on a host that is connected to a TCP/IP internet, what is its primary IP
address? Is the host connected to the worldwide Internet? Is it multihomed?

1.6 Obtain a copy of RFC 1000 to learn where the term RFC originated.

1.7 Contact the Internet Society, isoc@isoc.org or +1 703 648 9888, to find out about joining.

1.8 Fetch the fileabout-internic/information-about-the-internic using anonymous
FTP from the host is.internic.net.

29

Link Layer
2.1 Introduction
From Figure 1.4 we see that the purpose of the link layer in the TCP/IP protocol suite is to send and
receive (1) IP datagrams for the IP module, (2) ARP requests and replies for the ARP module, and (3)
RARP requests and replies for the RARP module. TCP/IP supports many different link layers,
depending on the type of networking hardware being used: Ethernet, token ring, FDDI (Fiber
Distributed Data Interface), RS-232 serial lines, and the like.

In this chapter we’ll look at some of the details involved in the Ethernet link layer, two specialized link
layers for serial interfaces (SLIP and PPP), and the loopback driver that’s part of most
implementations. Ethernet and SLIP are the link layers used for most of the examples in the book. We
also talk about the MTU (Maximum Transmission Unit), a characteristic of the link layer that we
encounter numerous times in the remaining chapters. We also show some calculations of how to
choose the MTU for a serial line.

2.2 Ethernet and IEEE 802 Encapsulation
The term Ethernet generally refers to a standard published in 1982 by Digital Equipment Corp., Intel
Corp., and Xerox Corp. It is the predominant form of local area network technology used with TCP/IP
today. It uses an access method called CSMA/CD, which stands for Carrier Sense, Multiple Access
with Collision Detection. It operates at 10 Mbits/sec and uses 48-bit addresses.

A few years later the IEEE (Institute of Electrical and Electronics Engineers) 802 Committee
published a sightly different set of standards. 802.3 covers an entire set of

CSMA/CD networks, 802.4 covers token bus networks, and 802.5 covers token ring networks.
Common to all three of these is the 802.2 standard that defines the logical link control (LLC) common
to many of the 802 networks. Unfortunately the combination of 802.2 and 802.3 defines a different
frame format from true Ethernet. ([Stallings 1987] covers all the details of these IEEE 802 standards.)

In the TCP/IP world, the encapsulation of IP datagrams is defined in RFC 894 [Hornig 1984] for
Ethernets and in RFC 1042 [Postel and Reynolds 1988] for IEEE 802 networks. The Host
Requirements RFC requires that every Internet host connected to a 10 Mbits/sec Ethernet cable:

1. Must be able to send and receive packets using RFC 894 (Ethernet) encapsulation.
2. Should be able to receive RFC 1042 (IEEE 802) packets intermixed with RFC 894 packets.
3. May be able to send packets using RFC 1042 encapsulation. If the host can send both types of
packets, the type of packet sent must be configurable and the configuration option must default to
RFC 894 packets.

RFC 894 encapsulation is most commonly used. Figure 2.1 shows the two different forms of
encapsulation. The number below each box in the figure is the size of that box in bytes.

Both frame formats use 48-bit (6-byte) destination and source addresses. (802.3 allows 16-bit
addresses to be used, but 48-bit addresses are normal.) These are what we call hardware addresses
throughout the text. The ARP and RARP protocols (Chapters 4 and 5) map between the 32-bit IP
addresses and the 48-bit hardware addresses.

30

The next 2 bytes are different in the two frame formats. The 802 length field says how many bytes
follow, up to but not including the CRC at the end. The Ethernet type field identifies the type of data
that follows. In the 802 frame the same type field occurs later in the SNAP (Sub-network Access
Protocol) header. Fortunately none of the valid 802 length values is the same as the Ethernet type
values, making the two frame formats distinguishable.

In the Ethernet frame the data immediately follows the type field, while in the 802 frame format 3
bytes of 802.2 LLC and 5 bytes of 802.2 SNAP follow. The DSAP (Destination Service Access Point)
and SSAP (Source Service Access Point) are both set to 0xaa. The Ctrl field is set to 3. The next 3
bytes, the org code are all 0. Following this is the same 2-byte type field that we had with the Ethernet
frame format. (Additional type field values are given in RFC 1340 [Reynolds and Postel 1992].)

The CRC field is a cyclic redundancy check (a checksum) that detects errors in the rest of the frame.
(This is also called the FCS or frame check sequence.)

There is a minimum size for 802.3 and Ethernet frames. This minimum requires that the data portion
be at least 38 bytes for 802.3 or 46 bytes for Ethernet. To handle this, pad bytes are inserted to assure
that the frame is long enough. We’ll encounter this minimum when we start watching packets on the
wire.

In this text we’ll display the Ethernet encapsulation when we need to, because this is the most
commonly used form of encapsulation.

31

Figure 2.1 IEEE 802.2/802.3 encapsulation (RFC 1042) and Ethernet encapsulation (RFC 894).

2.3 Trailer Encapsulation
RFC 893 [Leffler and Karels 1984] describes another form of encapsulation used on Ethernets, called
trailer encapsulation. It was an experiment with early BSD systems on DEC VAXes that improved
performance by rearranging the order of the fields in the IP datagram. "The variable-length fields at
the beginning of the data portion of the Ethernet frame (the IP header and the TCP header) were
moved to the end (right before the CRC). This allows the data portion of the frame to be mapped to a
hardware page, saving a memory-to-memory copy when the data is copied in the kernel. TCP data that
is a multiple of 512 bytes in size can be moved by just manipulating the kernel’s page tables. Two
hosts negotiated the use of trailer encapsulation using an extension of ARP. Different Ethernet frame
type values are defined for these frames.

Nowadays trailer encapsulation is deprecated, so we won’t show any examples of it. Interested readers
are referred to RFC 893 and Section 11.8 of [Leffler et al. 1989] for additional details.

32

2.4 SLIP: Serial Line IP
SLIP stands for Serial Line IP. It is a simple form of encapsulation for IP datagrams on serial lines,
and is specified in RFC 1055 [Rornkey 1988]. SLIP has become popular for connecting home systems
to the Internet, through the ubiquitous RS-232 serial port found on almost every computer and
high-speed modems. The following rules specify the framing used by SLIP.

1. The IP datagram is terminated by the special character called END (0xc0). Also, to prevent any
line noise before this datagram from being interpreted as part of this datagram, most
implementations transmit an END character at the beginning of the datagram too. (If there was
some line noise, this END terminates that erroneous datagram, allowing the current datagram to
be transmitted. That erroneous datagram will be thrown away by a higher layer when its contents
are detected to be garbage.)
2. If a byte of the IP datagram equals the END character, the 2-byte sequence 0xdb, 0xdc is
transmitted instead. This special character, 0xdb, is called the SLIP ESC character, but its value
is different from the ASCII ESC character (0xib).
3. If a byte of the IP datagram equals the SLIP ESC character, the 2-byte sequence 0xdb, 0xdd is
transmitted instead.

Figure 2.2 shows an example of this framing, assuming that one END character and one ESC character
appear in the original IP datagram. In this example the number of bytes transmitted across the serial
line is the length of the IP datagram plus 4.

Figure 2.2 SLIP encapsulation

SLIP is a simple framing method. It has some deficiencies that are worth noting.

1. Each end must know the other’s IP address. There is no method for one end to inform the other of
its IP address.
2. There is no type field (similar to the frame type field in Ethernet frames). If a serial line is used
for SLIP, it can’t be used for some other protocol at the same time.
3. There is no checksum added by SLIP (similar to the CRC field in Ethernet frames). If a noisy
phone line corrupts a datagram being transferred by SLIP, it’s up to the higher layers to detect
this. (Alternately, newer modems can detect and correct corrupted frames.) This makes it
essential that the upper layers provide some form of CRC. In Chapters 3 and 17 we’ll see that
there is always a checksum for the IP header, and for the TCP header and the TCP data. But in
Chapter 11 we’ll see that the checksum that covers the UDP header and UDP data is optional.

33

Despite these shortcomings, SLIP is a popular protocol that is widely used.

The history of SLIP dates back to 1984 when Rick Adams implemented it in 4.2BSD. Despite its self-description as a
nonstandard, it is becoming more popular as the speed and reliability of modems increase. Publicly available
implementations abound, and many vendors support it today.

2.5 Compressed SLIP
Since SLIP lines are often slow (19200 bits/sec or below) and frequently used for interactive traffic
(such as Telnet and Rlogin, both of which use TCP), there tend to be many small TCP packets
exchanged across a SLIP line. To carry I byte of data requires a 20-byte IP header and a 20-byte TCP
header, an overhead of 40 bytes. (Section 19.2 shows the flow of these small packets when a simple
command is typed during an Rlogin session.)

Recognizing this performance drawback, a newer version of SLIP, called CSLIP (for compressed
SLIP), is specified in RFC 1144 [Jacobson 1990a]. CSLIP normally reduces the 40-byte header to 3 or
5 bytes. It maintains the state of up to 16 TCP connections on each end of the CSLIP link and knows
that some of the fields in the two headers for a given connection normally don’t change. Of the fields
that do change, most change by a small positive amount. These smaller headers greatly improve the
interactive response time.

Most SLIP implementations today support CSLIP. Both SLIP links on the author’s subnet (see inside front cover) are
CSLIP links.

2.6 PPP: Point-to-Point Protocol
PPP, the Point-to-Point Protocol, corrects all the deficiencies in SLIP. PPP consists of three
components.

1. A way to encapsulate IP datagrams on a serial link. PPP supports either an asynchronous link
with 8 bits of data and no parity (i.e., the ubiquitous serial interface found on most computers) or
bit-oriented synchronous links.
2. A link control protocol (LCP) to establish, configure, and test the data-link connection. This
allows each end to negotiate various options.
3. A family of network control protocols (NCPs) specific to different network layer protocols. RFCs
currently exist for IP, the OSI network layer, DECnet, and AppleTalk. The IP NCP, for example,
allows each end to specify if it can perform header compression, similar to CSLIP. (The acronym
NCP was also used for the predecessor to TCP.)

RFC 1548 [Simpson 1993] specifies the encapsulation method and the link control protocol. RFC
1332 [McGregor 1992] specifies the network control protocol for IP.

The format of the PPP frames was chosen to look like the ISO HDLC standard (high-level data link
control). Figure 2.3 shows the format of PPP frames.

34

Figure 2.3 Format of PPP frames.

Each frame begins and ends with a flag byte whose value is 0x7e. This is followed by an address
byte whose value is always 0xff, and then a control byte, with a value of 0x03.

Next comes the protocol field, similar in function to the Ethernet type field. A value of 0x0021
means the information field is an IP datagram, a value of 0xc021 means the information field is link
control data, and a value of 0x8021 is for network control data.

The CRC field (or FCS, for frame check sequence) is a cyclic redundancy check, to detect errors in the
frame.

Since the byte value 0x7e is the flag character, PPP needs to escape this byte when it appears in the
information field. On a synchronous link this is done by the hardware using a technique called bit
stuffing [Tanenbaum 1989]. On asynchronous links the special byte 0x7d is used as an escape
character. Whenever this escape character appears in a PPP frame, the next character in the frame has
had its sixth bit complemented, as follows:

1. The byte 0x7e is transmitted as the 2-byte sequence 0x7d, 0x5e. This is the escape of the flag
byte.
2. The byte 0x7d is transmitted as the 2-byte sequence 0x7d, 0x5d. This is the escape of the
escape byte.
3. By default, a byte with a value less than 0x20 (i.e., an ASCII control character) is also escaped.
For example, the byte 0x01 is transmitted as the 2-byte sequence 0x7d, 0x21. (In this case the
complement of the sixth bit turns the bit on, whereas in the two previous examples the
complement turned the bit off.)

The reason for doing this is to prevent these bytes from appearing as ASCII control characters to
the serial driver on either host, or to the modems, which sometimes interpret these control
characters specially. It is also possible to use the link control protocol to specify which, if any, of
these 32 values must be escaped. By default, all 32 are escaped.

Since PPP, like SLIP, is often used across slow serial links, reducing the number of bytes per frame
reduces the latency for interactive applications. Using the link control protocol, most implementations
negotiate to omit the constant address and control fields and to reduce the size of the protocol field
from 2 bytes to 1 byte. If we then compare the framing overhead in a PPP frame, versus the 2-byte
framing overhead in a SLIP frame (Figure 2.2), we see that PPP adds three additional bytes: I byte for

35

the protocol field, and 2 bytes for the CRC. Additionally, using the IP network control protocol, most
implementations then negotiate to use Van Jacobson header compression (identical to CSLIP
compression) to reduce the size of the IP and TCP headers.

In summary, PPP provides the following advantages over SLIP: (1) support for multiple protocols on a
single serial line, not just IP datagrams, (2) a cyclic redundancy check on every frame, (3) dynamic
negotiation of the IP address for each end (using the IP network control protocol), (4) TCP and IP
header compression similar to CSLIP, and (5) a link control protocol for negotiating many data-link
options. The price we pay for all these features is 3 bytes of additional overhead per frame, a few
frames of negotiation when the link is established, and a more complex implementation.

Despite all the added benefits of PPP over SLIP, today there are more SLIP users than PPP users. As implementations
become more widely available, and as vendors start to support PPP, it should (eventually) replace SLIP.

2.7 Loopback Interface
Most implementations support a loopback interface that allows a client and server on the same host to
communicate with each other using TCP/IP. The class A network ID 127 is reserved for the loopback
interface. By convention, most systems assign the IP address of 127.0.0.1 to this interface and assign it
the name localhost. An IP datagram sent to the loopback interface must not appear on any
network.

Although we could imagine the transport layer detecting that the other end is the loopback address,
and short circuiting some of the transport layer logic and all of the network layer logic, most
implementations perform complete processing of the data in the transport layer and network layer, and
only loop the IP datagram back to itself when the datagram leaves the bottom of the network layer.

Figure 2.4 shows a simplified diagram of how the loopback interface processes IP datagrams.

36

Figure 2.4 Processing of IP datagrams by loopback interface.

The key points to note in this figure are as follows:

1. Everything sent to the loopback address (normally 127.0.0.1) appears as IP input.
2. Datagrams sent to a broadcast address or a multicast address are copied to the loopback interface
and sent out on the Ethernet. This is because the definition of broadcasting and multicasting
(Chapter 12) includes the sending host.
3. Anything sent to one of the host’s own IP addresses is sent to the loopback interface.

While it may seem inefficient to perform all the transport layer and IP layer processing of the loopback
data, it simplifies the design because the loopback interface appears as just another link layer to the
network layer. The network layer passes a datagram to the loopback interface like any other link layer,
and it happens that the loopback interface then puts the datagram back onto IP’s input queue.

Another implication of Figure 2.4 is that IP datagrams sent to the one of the host’s own IP addresses
normally do not appear on the corresponding network. For example, on an Ethernet, normally the
packet is not transmitted and then read back. Comments in some of the BSD Ethernet device drivers
indicate that many Ethernet interface cards are not capable of reading their own transmissions. Since a
host must process IP datagrams that it sends to itself, handling these packets as shown in Figure 2.4 is
the simplest way to accomplish this.

37

The 4.4BSD implementation defines the variable useloopback and initializes it to 1. If this variable is set to 0, however,
the Ethernet driver sends local packets onto the network instead of sending them to the loopback driver. This may or may
not work, depending on your Ethernet interface card and device driver.

2.8 MTU
As we can see from Figure 2.1, there is a limit on the size of the frame for both Ethernet encapsulation
and 802.3 encapsulation. This limits the number of bytes of data to 1500 and 1492, respectively. This
characteristic of the link layer is called the MTU, its maximum transmission unit. Most types of
networks have an upper limit.

If IP has a datagram to send, and the datagram is larger than the link layer’s MTU, IP performs
fragmentation, breaking the datagram up into smaller pieces (fragments), so that each fragment is
smaller than the MTU. We discuss IP fragmentation in Section 11.5.

Figure 2.5 lists some typical MTU values, taken from RFC 1191 [Mogul and Deering 1990]. The
listed MTU for a point-to-point link (e.g., SLIP or PPP) is not a physical characteristic of the network
media. Instead it is a logical limit to provide adequate response time for interactive use. In the Section
2.10 we’ll see where this limit comes from.

In Section 3.9 we’ll use the netstat command to print the MTU of an interface.

Network MTU (bytes)
Hyperchannel 65535
16 Mbits/sec token ring (IBM) 17914
4 Mbits/sec token ring (IEEE 802.5) 4464
FDDI 4352
Ethernet 1500
IEEE 802.3/802.2 1492
X.25 576
Point-to-Point (low delay) 296

Figure 2.5 Typical maximum transmission units (MTUs).

2.9 Path MTU
When two hosts on the same network are communicating with each other, it is the MTU of the
network that is important. But when two hosts are communicating across multiple networks, each link
can have a different MTU. The important numbers are not the MTUs of the two networks to which the
two hosts connect, but rather the smallest MTU of any data link that packets traverse between the two
hosts. This is called the path MTU.

The path MTU between any two hosts need not be constant. It depends on the route being used at any
time. Also, routing need not be symmetric (the route from A to B may not be the reverse of the route
from B to A), hence the path MTU need not be the same in the two directions.

RFC 1191 [Mogul and Deering 1990] specifies the "path MTU discovery mechanism," a way to
determine the path MTU at any time. We’ll see how this mechanism operates after we’ve described
ICMP and IP fragmentation. In Section 11.6 we’ll examine the ICMP unreachable error that is used
with this discovery mechanism and in Section 11.7 we’ll show a version of the traceroute

38

program that uses this mechanism to determine the path MTU to a destination. Sections 11.8 and 24.2
show how UDP and TCP operate when the implementation supports path MTU discovery.

2.10 Serial Line Throughput Calculations
If the line speed is 9600 bits/sec, with 8 bits per byte, plus I start bit and I stop bit, the line speed is 960
bytes/sec. Transferring a 1024-byte packet at this speed takes 1066 ms. If we’re using the SLIP link
for an interactive application, along with an application such as FTP that sends or receives 1024-byte
packets, we have to wait, on the average, half of this time (533 ms) to send our interactive packet.

This assumes that our interactive packet will be sent across the link before any further "big" packets.
Most SLIP implementations do provide this type-of-service queuing, placing interactive traffic ahead
of bulk data traffic. The interactive traffic is normally Telnet, Rlogin, and the control portion (the user
commands, not the data) of FTP.

This type of service queuing is imperfect. It cannot affect noninteractive traffic that is already queued
downstream (e.g., at the serial driver). Also newer modems have large buffers so noninteractive traffic
may already be buffered in the modem.

Waiting 533 ms is unacceptable for interactive response. Human factors studies have found that an
interactive response time longer than 100-200 ms is perceived as bad [Jacobson 1990a]. This is the
round-trip time for an interactive packet to be sent and something to be returned (normally a character
echo).

Reducing the MTU of the SLIP link to 256 means the maximum amount of time the link can be busy
with a single frame is 266 ms, and half of this (our average wait) is 133 ms. This is better, but still not
perfect. The reason we choose this value (as compared to 64 or 128) is to provide good utilization of
the line for bulk data transfers (such as large file transfers). Assuming a 5-byte CSLIP header, 256
bytes of data in a 261-byte frame gives 98.1% of the line to data and 1.9% to headers, which is good
utilization. Reducing the MTU below 256 reduces the maximum throughput that we can achieve for
bulk data transfers.

The MTU value listed in Figure 2.5, 296 for a point-to-point link, assumes 256 bytes of data and the
40-byte TCP and IP headers. Since the MTU is a value that IP queries the link layer for, the value must
include the normal TCP and IP headers. This is how IP makes its fragmentation decision. IP knows
nothing about the header compression that CSLIP performs.

Our average wait calculation (one-half the time required to transfer a maximum sized frame) only
applies when a SLIP link (or PPP link) is used for both interactive traffic and bulk data transfer. When
only interactive traffic is being exchanged, 1 byte of data in each direction (assuming 5-byte
compressed headers) takes around 12.5 ms for the round trip at 9600 bits/sec. This is well within the
100-200 ms range mentioned earlier. Also notice that compressing the headers from 40 bytes to 5
bytes reduces the round-trip time for the I byte of data from 85 to 12.5 ms.

Unfortunately these types of calculations are harder to make when newer error correcting, compressing
modems are being used. The compression employed by these modems reduces the number of bytes
sent across the wire, but the error correction may increase the amount of time to transfer these bytes.
Nevertheless, these calculations give us a starting point to make reasonable decisions.

In later chapters we’ll use these serial line calculations to verify some of the timings that we see when
watching packets go across a serial link.

39

2.11 Summary
This chapter has examined the lowest layer in the Internet protocol suite, the link layer. We looked at
the difference between Ethernet and IEEE 802.2/802.3 encapsulation, and the encapsulation used by
SLIP and PPP. Since both SLIP and PPP are often used on slow links, both provide a way to compress
the common fields that don’t often change. This provides better interactive response.

The loopback interface is provided by most implementations. Access to this interface is either through
the special loopback address, normally 127.0.0.1, or by sending IP datagrams to one of the host’s own
IP addresses. Loopback data has been completely processed by the transport layer and by IP when it
loops around to go up the protocol stack.

We described an important feature of many link layers, the MTU, and the related concept of a path
MTU. Using the typical MTUs for serial lines, we calculated the latency involved in SLIP and CSLIP
links.

This chapter has covered only a few of the common data-link technologies used with TCP/IP today.
One reason for the success of TCP/IP is its ability to work on top of almost any data-link technology.

Exercises
2.1 If your system supports the netstat(1) command (see Section 3.9 also), use it to determine the
interfaces on your system and their MTUs.

40

IP: Internet Protocol
3.1 Introduction
IP is the workhorse protocol of the TCP/IP protocol suite. All TCP, UDP, ICMP, and IGMP data gets
transmitted as IP datagrams (Figure 1.4). A fact that amazes many newcomers to TCP/IP, especially
those from an X.25 or SNA background, is that IP provides an unreliable, connectionless datagram
delivery service.

By unreliable we mean there are no guarantees that an IP datagram successfully gets to its destination.
IP provides a best effort service. When something goes wrong, such as a router temporarily running
out of buffers, IP has a simple error handling algorithm: throw away the datagram and try to send an
ICMP message back to the source. Any required reliability must be provided by the upper layers (e.g.,
TCP).

The term connectionless means that IP does not maintain any state information about successive
datagrams. Each datagram is handled independently from all other datagrams. This also means that IP
datagrams can get delivered out of order. If a source sends two consecutive datagrams (first A, then B)
to the same destination, each is routed independently and can take different routes, with B arriving
before A.

In this chapter we take a brief look at the fields in the IP header, describe IP routing, and cover
subnetting. We also look at two useful commands: ifconfig and netstat. We leave a detailed
discussion of some of the fields in the IP header for later when we can see exactly how the fields are
used. RFC 791 [Postel 1981a] is the official specification of IP.

3.2 IP Header
Figure 3.1 shows the format of an IP datagram. The normal size of the IP header is 20 bytes, unless
options are present.

41

Figure 3.1 IP datagram, showing the fields in the IP header.

We will show the pictures of protocol headers in TCP/IP as in Figure 3.1. The most significant bit is
numbered 0 at the left, and the least significant bit of a 32-bit value is numbered 31 on the right.

The 4 bytes in the 32-bit value are transmitted in the order: bits 0-7 first, then bits 8-15, then 16-23,
and bits 24-31 last. This is called big endian byte ordering, which is the byte ordering required for all
binary integers in the TCP/IP headers as they traverse a network. This is called the network byte order.
Machines that store binary integers in other formats, such as the little endian format, must convert the
header values into the network byte order before transmitting the data.

The current protocol version is 4, so IP is sometimes called IPv4. Section 3.10 discusses some
proposals for a new version of IP.

The header length is the number of 32-bit words in the header, including any options. Since this is a
4-bit field, it limits the header to 60 bytes. In Chapter 8 we’ll see that this limitation makes some of the
options, such as the record route option, useless today. The normal value of this field (when no options
are present) is 5.

The type-of-service field (TOS) is composed of a 3-bit precedence field (which is ignored today), 4
TOS bits, and an unused bit that must be 0. The 4 TOS bits are: minimize delay, maximize throughput,
maximize reliability, and minimize monetary cost.

Only 1 of these 4 bits can be turned on. If all 4 bits are 0 it implies normal service. RFC 1340
[Reynolds and Postel 1992] specifies how these bits should be set by all the standard applications.
RFC 1349 [Almquist 1992] contains some corrections to this RFC, and a more detailed description of
the TOS feature.

42

Figure 3.2 shows the recommended values of the TOS field for various applications. In the final
column we show the hexadecimal value, since that’s what we’ll see in the tcpdump output later in
the text.

Minimize
Minimize Maximize Maximize Hex
Application monetary
delay throughput reliability value
cost
Telnet/Rlogin 1 0 0 0 0x10
FTP
1 0 0 0 0x10
control
0 1 0 0 0x08
data
any bulk data 0 1 0 0 0x08
TFTP 1 0 0 0 0x10
SMTP
command 1 0 0 0 0x10
phase 0 1 0 0 0x08
data phase
DNS
1 0 0 0 0x10
UDP query
0 0 0 0 0x00
TCP query
0 1 0 0 0x08
zone transfer
ICMP
0 0 0 0 0x00
error
0 0 0 0 0x00
query
any IGP 0 0 1 0 0x04
SNMP 0 0 1 0 0x04
BOOTP 0 0 0 0 0x00
NNTP 0 0 0 1 0x02

Figure 3.2 Recommended values for type-of-service field.

The interactive login applications, Telnet and Rlogin, want a minimum delay since they’re used
interactively by a human for small amounts of data transfer. File transfer by FTP, on the other hand,
wants maximum throughput. Maximum reliability is specified for network management (SNMP) and
the routing protocols. Usenet news (NNTP) is the only one shown that wants to minimize monetary
cost.

The TOS feature is not supported by most TCP/IP implementations today, though newer systems
starting with 4.3BSD Reno are setting it. Additionally, new routing protocols such as OSPF and IS-IS
are capable of making routing decisions based on this field.

43

In Section 2.10 we mentioned that SLIP drivers normally provide type-of-service queuing, allowing interactive traffic to be
handled before bulk data. Since most implementations don’t use the TOS field, this queuing is done ad hoc by SLIP, with
the driver looking at the protocol field (to determine whether it’s a TCP segment or not) and then checking the source and
destination TCP port numbers to see if the port number corresponds to an interactive service. One driver comments that
this "disgusting hack" is required since most implementations don’t allow the application to set the TOS field.

The total length field is the total length of the IP datagram in bytes. Using this field and the header
length field, we know where the data portion of the IP datagram starts, and its length. Since this is a
16-bit field, the maximum size of an IP datagram is 65535 bytes. (Recall from Figure 2.5 that a
Hyperchannel has an MTU of 65535. This means there really isn’t an MTU-it uses the largest IP
datagram possible.) This field also changes when a datagram is fragmented, which we describe in
Section 11.5.

Although it’s possible to send a 65535-byte IP datagram, most link layers will fragment this.
Furthermore, a host is not required to receive a datagram larger than 576 bytes. TCP divides the user’s
data into pieces, so this limit normally doesn’t affect TCP. With UDP we’ll encounter numerous
applications in later chapters (RIP, TFTP, BOOTP, the DNS, and SNMP) that limit themselves to 512
bytes of user data, to stay below this 576-byte limit. Realistically, however, most implementations
today (especially those that support the Network File System, NFS) allow for just over 8192-byte IP
datagrams.

The total length field is required in the IP header since some data links (e.g., Ethernet) pad small
frames to be a minimum length. Even though the minimum Ethernet frame size is 46 bytes (Figure
2.1), an IP datagram can be smaller. If the total length field wasn’t provided, the IP layer wouldn’t
know how much of a 46-byte Ethernet frame was really an IP datagram.

The identification field uniquely identifies each datagram sent by a host. It normally increments by
one each time a datagram is sent. We return to this field when we look at fragmentation and
reassembly in Section 11.5. Similarly, we’ll also look at the flags field and the fragmentation offset
field when we talk about fragmentation.

RFC 791 [Postel 1981a] says that the identification field should be chosen by the upper layer that is having IP send the
datagram. This implies that two consecutive IP datagrams, one generated by TCP and one generated by UDP, can have the
same identification field. While this is OK (the reassembly algorithm handles this), most Berkeley-derived
implementations have the IP layer increment a kernel variable each time an IP datagram is sent, regardless of which layer
passed the data to IP to send. This kernel variable is initialized to a value based on the time-of-day when the system is
bootstrapped.

The time-to-live field, or TTL, sets an upper limit on the number of routers through which a datagram
can pass. It limits the lifetime of the datagram. It is initialized by the sender to some value (often 32 or
64) and decremented by one by every router that handles the datagram. When this field reaches 0, the
datagram is thrown away, and the sender is notified with an ICMP message. This prevents packets
from getting caught in routing loops forever. We return to this field in Chapter 8 when we look at the
Trace-route program.

We talked about the protocol field in Chapter 1 and showed how it is used by IP to demultiplex
incoming datagrams in Figure 1.8. It identifies which protocol gave the data for IP to send.

The header checksum is calculated over the IP header only. It does not cover any data that follows the
header. ICMP, IGMP, UDP, and TCP all have a checksum in their own headers to cover their header
and data.

44

To compute the IP checksum for an outgoing datagram, the value of the checksum field is first set to 0.
Then the 16-bit one’s complement sum of the header is calculated (i.e., the entire header is considered
a sequence of 16-bit words). The 16-bit one’s complement of this sum is stored in the checksum field.
When an IP datagram is received, the 16-bit one’s complement sum of the header is calculated. Since
the receiver’s calculated checksum contains the checksum stored by the sender, the receiver’s
checksum is all one bits if nothing in the header was modified. If the result is not all one bits (a
checksum error), IP discards the received datagram. No error message is generated. It is up to the
higher layers to somehow detect the missing datagram and retransmit.

ICMP, IGMP, UDP, and TCP all use the same checksum algorithm, although TCP and UDP include
various fields from the IP header, in addition to their own header and data. RFC 1071 [Braden,
Borman, and Partridge 1988] contains implementation techniques for computing the Internet
checksum. Since a router often changes only the TTL field (decrementing it by 1), a router can
incrementally update the checksum when it forwards a received datagram, instead of calculating the
checksum over the entire IP header again. RFC 1141 [Mallory and Kullberg 1990] describes an
efficient way to do this.

The standard BSD implementation, however, does not use this incremental update feature when forwarding a datagram.

Every IP datagram contains the source IP address and the destination IP address. These are the 32-bit
values that we described in Section 1.4.

The final field, the options, is a variable-length list of optional information for the datagram. The
options currently defined are:

security and handling restrictions (for military applications, refer to RFC 1108 [Kent 1991] for
details),
record route (have each router record its IP address. Section 7.3),
timestamp (have each router record its IP address and time. Section 7.4),
loose source routing (specifying a list of IP addresses that must be traversed by the datagram.
Section 8.5), and
strict source routing (similar to loose source routing but here only the addresses in the list can be
traversed. Section 8.5).

These options are rarely used and not all host and routers support all the options.

The options field always ends on a 32-bit boundary. Pad bytes with a value of 0 are added if necessary.
This assures that the IP header is always a multiple of 32 bits (as required for the header length field).

3.3 IP Routing
Conceptually, IP routing is simple, especially for a host. If the destination is directly connected to the
host (e.g., a point-to-point link) or on a shared network (e.g., Ethernet or token ring), then the IP
datagram is sent directly to the destination. Otherwise the host sends the datagram to a default router,
and lets the router deliver the datagram to its destination. This simple scheme handles most host
configurations.

In this section and in Chapter 9 we’ll look at the more general case where the IP layer can be
configured to act as a router in addition to acting as a host. Most multiuser systems today, including
almost every Unix system, can be configured to act as a router. We can then specify a single routing
algorithm that both hosts and routers can use. The fundamental difference is that a host never forwards
datagrams from one of its interfaces to another, while a router forwards datagrams. A host that

45

contains embedded router functionality should never forward a datagram unless it has been
specifically configured to do so. We say more about this configuration option in Section 9.4.

In our general scheme, IP can receive a datagram from TCP, UDP, ICMP, or IGMP (that is, a locally
generated datagram) to send, or one that has been received from a network interface (a datagram to
forward). The IP layer has a routing table in memory that it searches each time it receives a datagram
to send. When a datagram is received from a network interface, IP first checks if the destination IP
address is one of its own IP addresses or an IP broadcast address. If so, the datagram is delivered to the
protocol module specified by the protocol field in the IP header. If the datagram is not destined for this
IP layer, then (1) if the IP layer was configured to act as a router the packet is forwarded (that is,
handled as an outgoing datagram as described below), else (2) the datagram is silently discarded.

Each entry in the routing table contains the following information:

Destination IP address. This can be either a complete host address or a network address, as
specified by the flag field (described below) for this entry. A host address has a nonzero host ID
(Figure 1.5) and identifies one particular host, while a network address has a host ID of 0 and
identifies all the hosts on that network (e.g., Ethernet, token ring).
IP address of a next-hop router, or the IP address of a directly connected network. A next-hop
router is one that is on a directly connected network to which we can send datagrams for delivery.
The next-hop router is not the final destination, but it takes the datagrams we send it and forwards
them to the final destination.
Flags. One flag specifies whether the destination IP address is the address of a network or the
address of a host. Another flag says whether the next-hop router field is really a next-hop router
or a directly connected interface. (We describe each of these flags in Section 9.2.)
Specification of which network interface the datagram should be passed to for transmission.

IP routing is done on a hop-by-hop basis. As we can see from this routing table information, IP does
not know the complete route to any destination (except, of course, those destinations that are directly
connected to the sending host). All that IP routing provides is the IP address of the next-hop router to
which the datagram is sent. It is assumed that the next-hop router is really "closer" to the destination
than the sending host is, and that the next-hop router is directly connected to the sending host.

IP routing performs the following actions:

1. Search the routing table for an entry that matches the complete destination IP address (matching
network ID and host ID). If found, send the packet to the indicated next-hop router or to the
directly connected interface (depending on the flags field). Point-to-point links are found here, for
example, since the other end of such a link is the other host’s complete IP address.

2. Search the routing table for an entry that matches just the destination network ID. If found, send
the packet to the indicated next-hop router or to the directly connected interface (depending on
the flags field). All the hosts on the destination network can be handled with this single routing
table entry All the hosts on a local Ethernet, for example, are handled with a routing table entry of
this type.

This check for a network match must take into account a possible subnet mask, which we
describe in the next section.

46

3. Search the routing table for an entry labeled "default." If found, send the packet to the indicated
next-hop router.

If none of the steps works, the datagram is undeliverable. If the undeliverable datagram was generated
on this host, a "host unreachable" or "network unreachable" error is normally returned to the
application that generated the datagram.

A complete matching host address is searched for before a matching network ID. Only if both of these
fail is a default route used. Default routes, along with the ICMP redirect message sent by a next-hop
router (if we chose the wrong default for a datagram), are powerful features of IP routing that we’ll
come back to in Chapter 9.

The ability to specify a route to a network, and not have to specify a route to every host, is another
fundamental feature of IP routing. Doing this allows the routers on the Internet, for example, to have a
routing table with thousands of entries, instead of a routing table with more than one million entries.

Examples
First consider a simple example: our host bsdi has an IP datagram to send to our host sun. Both hosts
are on the same Ethernet (see inside front cover). Figure 3.3 shows the delivery of the datagram.

When IP receives the datagram from one of the upper layers it searches its routing table and finds that
the destination IP address (140.252.13.33) is on a directly connected network (the Ethernet
140.252.13.0). A matching network address is found in the routing table. (In the next section we’ll see
that because of subnetting the network address of this Ethernet is really 140.252.13.32, but that doesn’t
affect this discussion of routing.) The datagram is passed to the Ethernet device driver, and sent to sun
as an Ethernet frame (Figure 2.1). The destination address in the IP datagram is Sun’s IP address
(140.252.13.33) and the destination address in the link-layer header is the 48-bit Ethernet address of
sun’s Ethernet interface. This 48-bit Ethernet address is obtained using ARP, as we describe in the
next chapter. destination network = 140.252.13.0

Figure 3.3 Delivery of IP datagram from bsdi to sun.

Now consider another example: bsdi has an IP datagram to send to the host ftp.uu.net, whose IP
address is 192.48.96.9. Figure 3.4 shows the path of the datagram through the first three routers. First
bsdi searches its routing table but doesn’t find a matching host entry or a matching network entry. It
uses its default entry, which tells it to send datagrams to sun, the next-hop router. When the datagram
travels from bsdi to sun the destination IP address is the final destination (192.48.96.9) but the
link-layer address is the 48-bit Ethernet address of sun’s Ethernet interface. Compare this datagram
with the one in Figure 3.3, where the destination IP address and the destination link-layer address

47

specified the same host (sun).

When sun receives the datagram it realizes that the datagram’s destination IP address is not one of its
own, and sun is configured to act as a router, so it forwards the datagram. Its routing table is searched
and the default entry is used. The default entry on sun tells it to send datagrams to the next-hop router
netb, whose IP address is 140.252.1.183. The datagram is sent across the point-to-point SLIP link,
using the minimal encapsulation we showed in Figure 2.2. We don’t show a link-layer header, as we
do on the Ethernets, because there isn’t one on a SLIP link.

When netb receives the datagram it goes through the same steps that sun just did: the datagram is
not destined for one of its own IP addresses, and netb is configured to act as a router, so the datagram
is forwarded. The default routing table entry is used, sending the datagram to the next-hop router
gateway (140.252.1.4). ARP is used by netb on the Ethernet 140.252.1 to obtain the 48-bit
Ethernet address corresponding to 140.252.1.4, and that Ethernet address is the destination address in
the link-layer header.

gateway goes through the same steps as the previous two routers and its default routing table entry
specifies 140.252.104.2 as the next-hop router. (We’ll verify that this is the next-hop router for
gateway using Traceroute in Figure 8.4.)

Figure 3.4 Initial path of datagram from bsdi to ftp.uu.net (192.48.96.9).

48

A few key points come out in this example.

1. All the hosts and routers in this example used a default route. Indeed, most hosts and some
routers can use a default route for everything other than destinations on local networks.
2. The destination IP address in the datagram never changes. (In Section 8.5 we’ll see that this is not
true only if source routing is used, which is rare.) All the routing decisions are based on this
destination address.
3. A different link-layer header can be used on each link, and the link-layer destination address (if
present) always contains the link-layer address of the next hop. In our example both Ethernets
encapsulated a link-layer header containing the next-hop’s Ethernet address, but the SLIP link did
not. The Ethernet addresses are normally obtained using ARP.

In Chapter 9 we’ll look at IP routing again, after describing ICMP. We’ll also look at some sample
routing tables and how they’re used for routing decisions.

3.4 Subnet Addressing
All hosts are now required to support subnet addressing (RFC 950 [Mogul and Postel 1985]). Instead
of considering an IP address as just a network ID and host ID, the host ID portion is divided into a
subnet ID and a host ID.

This makes sense because class A and class B addresses have too many bits allocated for the host ID:
2 24 -2 and 2 16 -2, respectively. People don’t attach that many hosts to a single network. (Figure 1.5
shows the format of the different classes of IP addresses.) We subtract 2 in these expressions because
host IDs of all zero bits or all one bits are invalid.

After obtaining an IP network ID of a certain class from the InterNIC, it is up to the local system
administrator whether to subnet or not, and if so, how many bits to allocate to the subnet ID and host
ID. For example, the internet used in this text has a class B network address (140.252) and of the
remaining 16 bits, 8 are for the subnet ID and 8 for the host ID. This is shown in Figure 3.5.

16 bits 8 bits 8 bits
Class B netid = 140.252 subnetid hostid

Figure 3.5 Subnetting a class B address.

This division allows 254 subnets, with 254 hosts per subnet.

Many administrators use the natural 8-bit boundary in the 16 bits of a class B host ID as the subnet
boundary. This makes it easier to determine the subnet ID from a dotted-decimal number, but there is
no requirement that the subnet boundary for a class A or class B address be on a byte boundary.

Most examples of subnetting describe it using a class B address. Subnetting is also allowed for a class
C address, but there are fewer bits to work with. Subnetting is rarely shown with a class A address
because there are so few class A addresses. (Most class A addresses are, however, subnetted.)

Subnetting hides the details of internal network organization (within a company or campus) to external
routers. Using our example network, all IP addresses have the class B network ID of 140.252. But
there are more than 30 subnets and more than 400 hosts distributed over those subnets. A single router
provides the connection to the Internet, as shown in Figure 3.6.

49

In this figure we have labeled most of the routers as Rn, where n is the subnet number. We show the
routers that connect these subnets, along with the nine systems from the figure on the inside front
cover. The Ethernets are shown as thicker lines, and the point-to-point links as dashed lines. We do not
show all the hosts on the various subnets. For example, there are more than 50 hosts on the 140.252.3
subnet, and more than 100 on the 140.252.1 subnet.

The advantage to using a single class B address with 30 subnets, compared to 30 class C addresses, is
that subnetting reduces the size of the Internet’s routing tables. The fact that the class B address
140.252 is subnetted is transparent to all Internet routers other than the ones within the 140.252 subnet.
To reach any host whose IP

Figure 3.6 Arrangement of most of the noao.edu 140.252 subnets.

address begins with 140.252, the external routers only need to know the path to the IP address
140.252.104.1. This means that only one routing table entry is needed for all the 140.252 networks,
instead of 30 entries if 30 class C addresses were used. Subnetting, therefore, reduces the size of
routing tables. (In Section 10.8 we’ll look at a new technique that helps reduce the size of routing
tables even if class C addresses are used.)

To show that subnetting is not transparent to routers within the subnet, assume in Figure 3.6 that a
datagram arrives at gateway from the Internet with a destination address of 140.252.57.1. The router
gateway needs to know that the subnet number is 57, and that datagrams for this subnet are sent to
kpno. Similarly kpno must send the datagram to R55, who then sends it to R57.

50

3.5 Subnet Mask
Part of the configuration of any host that takes place at bootstrap time is the specification of the host’s
IP address. Most systems have this stored in a disk file that’s read at bootstrap time, and we’ll see in
Chapter 5 how a diskless system can also find out its IP address when it’s bootstrapped.

In addition to the IP address, a host also needs to know how many bits are to be used for the subnet ID
and how many bits are for the host ID. This is also specified at bootstrap time using a subnet mask.
This mask is a 32-bit value containing one bits for the network ID and subnet ID, and zero bits for the
host ID. Figure 3.7 shows the formation of the subnet mask for two different partitions of a class B
address. The top example is the partitioning used at noao.edu, shown in Figure 3.5, where the
subnet ID and host ID are both 8 bits wide. The lower example shows a class B address partitioned for
a 10-bit subnet ID and a 6-bit host ID.

Figure 3.7 Example subnet mask for two different class B subnet arrangements.

Although IP addresses are normally written in dotted-decimal notation, subnet masks are often written
in hexadecimal, especially if the boundary is not a byte boundary, since the subnet mask is a bit mask.

Given its own IP address and its subnet mask, a host can determine if an IP datagram is destined for
(1) a host on its own subnet, (2) a host on a different subnet on its own network, or (3) a host on a
different network. Knowing your own IP address tells you whether you have a class A, B, or C address
(from the high-order bits), which tells you where the boundary is between the network ID and the
subnet ID. The subnet mask then tells you where the boundary is between the subnet ID and the host
ID.

Example
Assume our host address is 140.252.1.1 (a class B address) and our subnet mask is 255.255.255.0 (8
bits for the subnet ID and 8 bits for the host ID).

If a destination IP address is 140.252.4.5, we know that the class B network IDs are the same
(140.252), but the subnet IDs are different (1 and 4). Figure 3.8 shows how this comparison of
two IP addresses is done, using the subnet mask.
If the destination IP address is 140.252.1.22, the class B network IDs are the same (140.252), and
the subnet IDs are the same (1). The host IDs, however, are different.
If the destination IP address is 192.43.235.6 (a class C address), the network IDs are different. No
further comparisons can be made against this address.

51

Figure 3.8 Comparison of two class B addresses using a subnet mask.

The IP routing function makes comparisons like this all the time, given two IP addresses and a subnet
mask.

3.6 Special Case IP Addresses
Having described subnetting we now show the seven special case IP addresses in Figure 3.9. In this
figure, 0 means a field of all zero bits, -1 means a field of all one bits, and netid, subnetid, and hostid
mean the corresponding field that is neither all zero bits nor all one bits. A blank subnet ID column
means the address is not subnetted.

IP address Can appear as Description
net subnet
host ID source? destination?
ID ID
this host on this net (see restrictions below)
0 0 OK never
specified host on this net (see restrictions
0 hostid OK never
below)
127 anything OK OK loopback address (Section 2.7)
-1 -1 never OK limited broadcast (never forwarded)
netid subnetid -1 never OK net-directed broadcast to netid
netid 1 -1 never OK subnet-directed broadcast to netid, subnetid
netid -1 never OK all-subnets-directed broadcast to netid

Figure 3.9 Special case IP addresses.

We have divided this table into three sections. The first two entries are special case source addresses,
the next one is the special loopback address, and the final four are the broadcast addresses.

"The first two entries in the table, with a network ID of 0, can only appear as the source address as part
of an initialization procedure when a host is determining its own IP address, for example, when the
BOOTP protocol is being used (Chapter 16). In Section 12.2 we’ll examine the four types of broadcast
addresses in more detail.

52

3.7 A Subnet Example
This example shows the subnet used in the text, and how two different subnet masks are used. Figure
3.10 shows the arrangement.

Figure 3.10 Arrangement of hosts and networks for author’s subnet.

If you compare this figure with the one on the inside front cover, you’ll notice that we’ve omitted the
detail that the connection from the router sun to the top Ethernet in Figure 3.10 is really a dialup SLIP
connection. This detail doesn’t affect our description of subnetting in this section. We’ll return to this
detail in Section 4.6 when we describe proxy ARP.

The problem is that we have two separate networks within subnet 13: an Ethernet and a point-to-point
link (the hardwired SLIP link). (Point-to-point links always cause problems since each end normally
requires an IP address.) There could be more hosts and networks in the future, but not enough hosts
across the different networks to justify using another subnet number. Our solution is to extend the
subnet ID from 8 to II bits, and decrease the host ID from 8 to 5 bits. This is called variable-length
subnets since most networks within the 140.252 network use an 8-bit subnet mask while our network
uses an 11-bit subnet mask.

RFC 1009 [Braden and Postel 1987] allows a subnetted network to use more than one subnet mask. The new Router
Requirements RFC [Almquist 1993] requires support for this.

The problem, however, is that not all routing protocols exchange the subnet mask along with the destination network ID.
We’ll see in Chapter 10 that RIP does not support variable-length subnets, while RIP Version 2 and OSPF do. We don’t
have a problem with our example, since RIP isn’t required on the author’s subnet.

Figure 3.11 shows the IP address structure used within the author’s subnet. The first 8 bits of the
11-bit subnet ID are always 13 within the author’s subnet. For the remaining 3 bits of the subnet ID,
we use binary 001 for the Ethernet, and binary 010 for

53

Figure 3.11 Using variable-length subnets.

the point-to-point SLIP link. This variable-length subnet mask does not cause a problem for other
hosts and routers in the 140.252 network-as long as all datagrams destined for the subnet 140.252.13
are sent to the router sun (IP address 140.252.1.29) in Figure 3.10, and if sun knows about the 11-bit
subnet ID for the hosts on its subnet 13, everything is fine.

The subnet mask for all the interfaces on the 140.252.13 subnet is 255.255.255.224, or 0xffffffe0.
This indicates that the rightmost 5 bits are for the host ID, and the 27 bits to the left are the network ID
and subnet ID.

Figure 3.12 shows the allocation of IP addresses and subnet masks for the interfaces shown in Figure
3.10.

Host
Host IP address Subnet mask Net ID/Subnet ID Comment
ID
on subnet 1
140.252.1.29 255.255.255.0 140.252.1 29
sun on author’s
140.252.13.33 255.255.255.224 140.252.13.32 1
Ethernet
svr4 140.252.13.34 255.255.255.224 140.252.13.32 2
140.252.13.35 255.255.255.224 140.252.13.32 3 on Ethernet
bsdi
140.252.13.66 255.255.255.224 140.252.13.64 2 point-to-point
slip 140.252.13.65 255.255.255.224 140.252.13.64 1 point-to-point
broadcast addr on
140.252.13.63 255.255.255.224 140.252.13.32 31
Ethernet

Figure 3.12 IP addresses on author’s subnet.

The first column is labeled "Host," but both sun and bsdi also act as routers, since they are multihomed
and route packets from one interface to another.

The final row in this table notes that the broadcast address for the Ethernet in Figure 3.10 is
140.252.13.63: it is formed from the subnet ID of the Ethernet (140.252.13.32) and the low-order 5
bits in Figure 3.11 set to 1 (16+8+4+2+1 = 31). (We’ll see in Chapter 12 that this address is called the
subnet-directed broadcast address.)

54

3.8 ifconfig Command
Now that we’ve described the link layer and the IP layer we can show the command used to configure
or query a network interface for use by TCP/IP. The ifconfig(8) command is normally run at
bootstrap time to configure each interface on a host.

For dialup interfaces that may go up and down (such as SLIP links), ifconfig must be run (somehow)
each time the line is brought up or down. How this is done each time the SLIP link is brought up or
down depends on the SLIP software being used.

The following output shows the values for the author’s subnet. Compare these values with the values
in Figure 3.12.

sun % /usr/etc/ifconfig -a SunOS -a option says report on all interfaces
leO : flags=63<UP, BROADCAST, NOTRAILERS, RUNNING>
inet 140.252.13.33 netmask ffffffe0 broadcast 140.252.13.63
slO : flags=105KUP, POINTOPOINT, RUNNING, LINKO>
inet 140.252.1.29 -> 140.252.1.183 netmask ffffff00
loO: flags=49<UP,LOOPBACK,RUNNING>
inet 127.0.0.1 netmask ff000000

The loopback interface (Section 2.7) is considered a network interface. Its class A address is not
subnetted.

Other things to notice are that trailer encapsulation (Section 2.3) is not used on the Ethernet, and that
the Ethernet is capable of broadcasting, while the SLIP link is a point-to-point link.

The flag LINK0 for the SLIP interface is the configuration option that enables compressed slip
(CSLIP, Section 2.5). Other possible options are LINK1, which enables CSLIP if a compressed packet
is received from the other end, and LINK2, which causes all outgoing ICMP packets to be thrown
away. We’ll look at the destination address of this SLIP link in Section 4.6.

A comment in the installation instructions gives the reason for this last option: "This shouldn’t have to
be set, but some cretin pinging you can drive your throughput to zero."

bsdi is the other router. Since the -a option is a SunOS feature, we have to execute ifconfig
multiple times, specifying the interface name as an argument:

bsdi % /sbin/ifconfig weO
we0: flags=863<UP, BROADCAST, NOTRAILERS, RUNNING, SIMPLEX>
bsdi % /sbin/ifconfig slO
sl0 : flags=1011<UP, POINTOPOINT, LINKO
inet 140.252.13.66 -> 140.252.13.65 netmask ffffffe0

55

Here we see a new option for the Ethernet interface (we0): SIMPLEX. This 4.4BSD flag specifies that
the interface can’t hear its own transmissions. It is set in BSD/386 for all the Ethernet interfaces. When
set, if the interface is sending a frame to the broadcast address, a copy is made for the local host and
sent to the loopback address. (We show an example of this feature in Section 6.3.)

On the host slip the configuration of the SLIP interface is nearly identical to the output shown above
on bsdi, with the exception that the IP addresses of the two ends are swapped:

slip % /sbin/ifconfig slO
sl0 : flags=1011<UP, POINTOPOINT, LINK0
inet 140.252.13.65 --> 140.252.13.66 netmask ffffffe0

The final interface is the Ethernet interface on the host svr4. It is similar to the Ethernet output
shown earlier, except that SVR4’s version of ifconfigdoesn’t print the RUNNING flag:

svr4 % /usr/sbin/ifconfig emdO
emdO: flags=23<UP, BROADCAST, NOTRAILERS>

The ifconfig command normally supports other protocol families (other than TCP/IP) and has
numerous additional options. Check your system’s manual for these details.

3.9 netstat Command
The netstat(l) command also provides information about the interfaces on a system. The -i flag
prints the interface information, and the -n flag prints IP addresses instead of hostnames.

sun % netstat -in
Name Mtu Net/Dest Address lpkts lerrs Opkts Oerrs Collis Queue
leO 1500 140.252.13.32 140.252.13.33 67719 0 92133 0 1 0
slO 552 140.252.1.183 140.252.1.29 48035 0 54963 0 0 0
loO 1536 127.0.0.0 127.0.0.1 15548 0 15548 0 0 0

This command prints the MTU of each interface, the number of input packets, input errors, output
packets, output errors, collisions, and the current size of the output queue.

We’ll return to the netstat command in Chapter 9 when we use it to examine the routing table, and
in Chapter 13 when we use a modified version to see active multicast groups.

3.10 IP Futures
There are three problems with IP. They are a result of the phenomenal growth of the Internet over the
past few years. (See Exercise 1.2 also.)

56

1. Over half of all class B addresses have already been allocated. Current estimates predict
exhaustion of the class B address space around 1995, if they continue to be allocated as they have
been in the past.
2. 32-bit IP addresses in general are inadequate for the predicted long-term growth of the Internet.
3. The current routing structure is not hierarchical, but flat, requiring one routing table entry per
network. As the number of networks grows, amplified by the allocation of multiple class C
addresses to a site with multiple networks, instead of a single class B address, the size of the
routing tables grows.

CIDR (Classless Interdomain Routing) proposes a fix to the third problem that will extend the
usefulness of the current version of IP (IP version 4) into the next century. We discuss it in more detail
in Section 10.8.

Four proposals have been made for a new version of IP, often called IPng, for the next generation of
IP. The May 1993 issue of IEEE Network (vol. 7, no. 3) contains overviews of the first three
proposals, along with an article on CIDR. RFC 1454 [Dixon 1993] also compares the first three
proposals.

1. SIP, the Simple Internet Protocol. It proposes a minimal set of changes to IP that uses 64-bit
addresses and a different header format. (The first 4 bits of the header still contain the version
number, with a value other than 4.)
2. PIP. This proposal also uses larger, variable-length, hierarchical addresses with a different header
format.
3. TUBA, which stands for "TCP and UDP with Bigger Addresses," is based on the OSI CLNP
(Connectionless Network Protocol), an OSI protocol similar to IP. It provides much larger
addresses: variable length, up to 20 bytes. Since CLNP is an existing protocol, whereas SIP and
PIP are just proposals, documentation already exists on CLNP. RFC 1347 [Gallon 1992] provides
details on TUBA. Chapter 7 of [Periman 1992] contains a comparison of IPv4 and CLNP. Many
routers already support CLNP, but few hosts do.
4. TP/IX, which is described in RFC 1475 [Ullmann 1993]. As with SIP, it uses 64 bits for IP
addresses, but it also changes the TCP and UDP headers: 32-bit port number for both protocols,
along with 64-bit sequence numbers, 64-bit acknowledgment numbers, and 32-bit windows for
TCP.

The first three proposals use basically the same versions of TCP and UDP as the transport layers.

Since only one of these four proposals will be chosen as the successor to IPv4, and since the decision
may have been made by the time you read this, we won’t say any more about them. With the
forthcoming implementation of CIDR to handle the short-term problem, it will take many years to
implement the successor to IPv4.

3.11 Summary
We started this chapter with a description of the IP header and briefly described all the fields in this
header. We also gave an introduction to IP routing, and saw that host routing can be simple: the
destination is either on a directly connected network, in which case the datagram is sent directly to the
destination, or a default router is chosen.

Hosts and routers have a routing table that is used for all routing decisions. There are three types of
routes in the table: host specific, network specific, and optional default routes. There is a priority to the
entries in a routing table. A host route will be chosen over a network router, and a default route is used

57

only when no other route exists to the destination.

IP routing is done on a hop-by-hop basis. The destination IP address never changes as the datagram
proceeds through all the hops, but the encapsulation and destination link-layer address can change on
each hop. Most hosts and many routers use a default next-hop router for all nonlocal traffic.

Class A and B addresses are normally subnetted. The number of bits used for the subnet ID is
specified by the subnet mask. We gave a detailed example of this, using the author’s subnet, and
introduced variable-length subnets. The use of subnetting reduces the size of the Internet routing
tables, since many networks can often be accessed through a single point. Information on the
interfaces and networks is available through the ifconfig and netstat commands. This includes
the IP address of the interface, its subnet mask, broadcast address, and MTU.

We finished the chapter with a discussion of potential changes to the Internet protocol suite-the next
generation of IP.

Exercises
3.1 Must the loopback address be 127.0.0.1?

3.2 Identify the routers in Figure 3.6 with more than two network interfaces.

3.3 What’s the difference in the subnet mask for a class A address with 16 bits for the subnet ID and a
class B address with 8 bits for the subnet ID?

3.4 Read RFC 1219 [Tsuchiya 1991] for a recommended technique for assigning subnet IDs and host
IDs.

3.5 Is the subnet mask 255.255.0.255 valid for a class A address?

3.6 Why do you think the MTU of the loopback interface printed in Section 3.9 is set to 1536?

3.7 The TCP/IP protocol suite is built on a datagram network technology, the IP layer. Other
protocol suites are built on a connection-oriented network technology. Read [dark 1988] to
discover the three advantages the datagram network layer provides.

58

ARP: Address Resolution Protocol
4.1 Introduction
The problem that we deal with in this chapter is that IP addresses only make sense to the TCP/IP
protocol suite. A data link such as an Ethernet or a token ring has its own addressing scheme (often
48-bit addresses) to which any network layer using the data link must conform. A network such as an
Ethernet can be used by different network layers at the same time. For example, a collection of hosts
using TCP/IP and another collection of hosts using some PC network software can share the same
physical cable.

When an Ethernet frame is sent from one host on a LAN to another, it is the 48-bit Ethernet address
that determines for which interface the frame is destined. The device driver software never looks at the
destination IP address in the IP datagram.

Address resolution provides a mapping between the two different forms of addresses: 32-bit IP
addresses and whatever type of address the data link uses. RFC 826 [Plummer 1982] is the
specification of ARP.

Figure 4.1 shows the two protocols we talk about in this chapter and the next: ARP (address resolution
protocol) and RARP (reverse address resolution protocol).

Figure 4.1 Address resolution protocols: ARP and RARP.

ARP provides a dynamic mapping from an IP address to the corresponding hardware address. We use
the term dynamic since it happens automatically and is normally not a concern of either the application
user or the system administrator.

RARP is used by systems without a disk drive (normally diskless workstations or X terminals) but
requires manual configuration by the system administrator. We describe it in Chapter 5.

4.2 An Example
Whenever we type a command of the form

% ftp bsdi

the following steps take place. These numbered steps are shown in Figure 4.2.

1. The application, the FTP client, calls the function gethostbyname(3) to convert the hostname
(bsdi) into its 32-bit IP address. This function is called a resolverin the DNS (Domain Name
System), which we describe in Chapter 14. This conversion is done using the DNS, or on smaller
networks, a static hosts file (/etc/hosts).
2. The FTP client asks its TCP to establish a connection with that IP address.

59

3. TCP sends a connection request segment to the remote host by sending an IP datagram to its IP
address. (We’ll see the details of how this is done in Chapter 18.)
4. If the destination host is on a locally attached network (e.g., Ethernet, token ring, or the other end
of a point-to-point link), the IP datagram can be sent directly to that host. If the destination host is on a
remote network, the IP routing function determines the Internet address of a locally attached next-hop
router to send the IP datagram to. In either case the IP datagram is sent to a host or router on a locally
attached network.
5. Assuming an Ethernet, the sending host must convert the 32-bit IP address into a 48-bit Ethernet
address. A translation is required from the logical Internet address to its corresponding physical
hardware address. This is the function of ARP.

ARP is intended for broadcast networks where many hosts or routers are connected to a single
network.
6. ARP sends an Ethernet frame called an ARP request to every host on the network. This is called a
broadcast. We show the broadcast in Figure 4.2 with dashed lines. The ARP request contains the
IP address of the destination host (whose name is bsdi) and is the request "if you are the owner
of this IP address, please respond to me with your hardware address."

Figure 4.2 Operation of ARP when user types "ftp hostname".

7. The destination host’s ARP layer receives this broadcast, recognizes that the sender is asking for
its hardware address, and replies with an ARP reply. This reply contains the IP address and the
corresponding hardware address.
8. The ARP reply is received and the IP datagram that forced the ARP request-reply to be
exchanged can now be sent.
9. The IP datagram is sent to the destination host.

The fundamental concept behind ARP is that the network interface has a hardware address (a 48-bit
value for an Ethernet or token ring interface). Frames exchanged at the hardware level must be
addressed to the correct interface. But TCP/IP works with its own addresses: 32-bit IP addresses.
Knowing a host’s IP address doesn’t let the kernel send a frame to that host. The kernel (i.e., the
Ethernet driver) must know the destination’s hardware address to send it data. The function of ARP is
to provide a dynamic mapping between 32-bit IP addresses and the hardware addresses used by
various network technologies.

Point-to-point links don’t use ARP. When these links are configured (normally at bootstrap time) the
kernel must be told of the IP address at each end of the link. Hardware addresses such as Ethernet
addresses are not involved.

4.3 ARP Cache
Essential to the efficient operation of ARP is the maintenance of an ARP cache on each host. This
cache maintains the recent mappings from Internet addresses to hardware addresses. The normal
expiration time of an entry in the cache is 20 minutes from the time the entry was created.

60

We can examine the ARP cache with the arp(8) command. The -a option displays all entries in the
cache:

bsdi % arp -a
sun (140.252.13.33) at 8:0:20:3:f6:42
svr4 (140.252.13.34) at 0:0:c0:c2:9b:26

The 48-bit Ethernet addresses are displayed as six hexadecimal numbers separated by colons. We
discuss additional features of the arp command in Section 4.8.

4.4 ARP Packet Format
Figure 4.3 shows the format of an ARP request and an ARP reply packet, when used on an Ethernet to
resolve an IP address. (ARP is general enough to be used on other networks and can resolve addresses
other than IP addresses. The first four fields following the frame type field specify the types and sizes
of the final four fields.)

Figure 4.3 Format of ARP request or reply packet when used on an Ethernet.

The first two fields in the Ethernet header are the source and destination Ethernet addresses. The
special Ethernet destination address of all one bits means the broadcast address. All Ethernet interfaces
on the cable receive these frames.

The 2-byte Ethernet frame type specifies the type of data that follows. For an ARP request or an ARP
reply, this field is 0x0806.

The adjectives hardware and protocol are used to describe the fields in the ARP packets. For example,
an ARP request asks for the protocol address (an IP address in this case) corresponding to a hardware
address (an Ethernet address in this case).

The hard type field specifies the type of hardware address. Its value is 1 for an Ethernet. Prot type
specifies the type of protocol address being mapped. Its value is 0x0800 for IP addresses. This is
purposely the same value as the type field of an Ethernet frame containing an IP datagram. (See Figure
2.1.)

The next two 1-byte fields, hard size and prot size, specify the sizes in bytes of the hardware addresses
and the protocol addresses. For an ARP request or reply for an IP address on an Ethernet they are 6
and 4, respectively.

The op field specifies whether the operation is an ARP request (a value of 1), ARP reply (2), RARP
request (3), or RARP reply (4). (We talk about RARP in Chapter 5.) This field is required since the
frame type field is the same for an ARP request and an ARP reply

61

The next four fields that follow are the sender’s hardware address (an Ethernet address in this
example), the sender’s protocol address (an IP address), the target hardware address, and the target
protocol address. Notice there is some duplication of information: the sender’s hardware address is
available both in the Ethernet header and in the ARP request.

For an ARP request all the fields are filled in except the target hardware address. When a system
receives an ARP request directed to it, it fills in its hardware address, swaps the two sender addresses
with the two target addresses, sets the op field to 2, and sends the reply.

4.5 ARP Examples
In this section we’ll use the tcpdump command to see what really happens with ARP when we execute
normal TCP utilities such as Telnet. Appendix A contains additional details on the tcpdump
program.

Normal Example
To see the operation of ARP we’ll execute the telnet command, connecting to the discard server.

bsdi % arp -a verify ARP cache is empty
bsdi % telnet svr4
discard
Trying 140.252.13.34...
connect to the discard server
Connected to svr4.
Escape character is
’^]’.
^]
telnet> quit type Control, right bracket to get Telnet client prompt and
Connection closed. terminate

While this is happening we run the tcpdump command on another system (sun) with the -e option.
This displays the hardware addresses (which in our examples are 48-bit Ethernet addresses).

62

0:0:c0:6f:2d:40 ff:ff:ff:ff:ff:ff arp 60:
1 0.0
arp who-has svr4 tell bsdi
0.002174 0:0:c0:c2:9b:26 0:0:c0:6f:2d:40 arp 60:
2
(0.0022) arp reply svr4 is-at 0:0:c0:c2:9b:26
0:0:c0:6f:2d:40 0:0:c0:c2:9b:26 ip 60:
0.002831 bsdi.1030 > svr4.discard: S
3
(0.0007) 596459521:596459521(0)
win 4096 <mss 1024> [tos 0x10]
0:0:c0:c2:9b:26 0:0:c0:6f:2d:40 ip 60:
0.007834 svr4.discard > bsdi.1030: S
4
(0.0050) 3562228225:3562228225(0)
ack 596459522 win 4096 <mss 1024>
0:0:c0:6f:2d:40 0:0:c0:c2:9b:26 ip 60:
0.009615
5 bsdi.1030 > svr4.discard: . ack 1 win 4096
(0.0018)
[tos 0x10]

Figure 4.4 ARP request and ARP reply generated by TCP connection request.

Figure A.3 in Appendix A contains the raw output from tcpdump used for Figure 4.4. Since this is
the first example of tcpdump output in the text, you should review that appendix to see how we’ve
beautified the output.

We have deleted the final four lines of the tcpdump output that correspond to the termination of the
connection (which we cover in Chapter 18), since they’re not relevant to the discussion here.

In line 1 the hardware address of the source (bsdi) is 0:0:c0:6f:2d:40. The destination
hardware address is ff:ff:ff:ff:ff:ff, which is the Ethernet broadcast address. Every Ethernet
interface on the cable will receive the frame and process it, as shown in Figure 4.2.

The next output field on line 1, arp, means the frame type field is 0x0806, specifying either an ARP
request or an ARP reply.

The value 60 printed after the words arp and ip on each of the five lines is the length of the Ethernet
frame. Since the size of an ARP request and ARP reply is 42 bytes (28 bytes for the ARP message, 14
bytes for the Ethernet header), each frame has been padded to the Ethernet minimum: 60 bytes.

Referring to Figure 1.7, this minimum of 60 bytes starts with and includes the 14-byte Ethernet header,
but does not include the 4-byte Ethernet trailer. Some books state the minimum as 64 bytes, which
includes the Ethernet trailer. We purposely did not include the 14-byte Ethernet header in the
minimum of 46 bytes shown in Figure 1.7, since the corresponding maximum (1500 bytes) is what’s
referred to as the MTU-maximum transmission unit (Figure 2.5). We use the MTU often, because it
limits the size of an IP datagram, but are normally not concerned with the minimum. Most device
drivers or interface cards automatically pad an Ethernet frame to the minimum size. The IP datagrams
on lines 3,4, and 5 (containing the TCP segments) are all smaller than the minimum, and have also
been padded to 60 bytes.

63

The next field on line 1, arp who-has, identifies the frame as an ARP request with the IP address
of svr4 as the target IP address and the IP address of bsdi as the sender IP address, tcpdump prints
the hostnames corresponding to the IP address by default. (We’ll use the -n option in Section 4.7 to
see the actual IP addresses in an ARP request.)

From line 2 we see that while the ARP request is broadcast, the destination address of the ARP reply is
bsdi (0:0:c0:6f:2d:40). The ARP reply is sent directly to the requesting host; it is not
broadcast.

tcpdump prints arp reply for this frame, along with the hostname and hardware address of the
responder.

Line 3 is the first TCP segment requesting that a connection be established. Its destination hardware
address is the destination host (svr4). We’ll cover the details of this segment in Chapter 18.

The number printed after the line number on each line is the time (in seconds) when the packet was
received by tcpdump. Each line other than the first also contains the time difference (in seconds)
from the previous line, in parentheses. We can see in this figure that the time between sending the
ARP request and receiving the ARP reply is 2.2 ms. The first TCP segment is sent 0.7 ms after this.
The overhead involved in using ARP for dynamic address resolution in this example is less than 3 ms.

A final point from the tcpdump output is that we don’t see an ARP request from svr4 before it
sends its first TCP segment (line 4). While it’s possible that svr4 already had an entry for bsdi in its
ARP cache, normally when a system receives an ARP request, in addition to sending the ARP reply it
also saves the requestor’s hardware address and IP address in its own ARP cache. This is on the
logical assumption that if the requestor is about to send it an IP datagram, the receiver of the datagram
will probably send a reply.

ARP Request to a Nonexistent Host
What happens if the host being queried for is down or nonexistent? To see this we specify a
nonexistent Internet address-the network ID and subnet ID are that of the local Ethernet, but there is no
host with the specified host ID. From Figure 3.10 we see the host IDs 36 through 62 are nonexistent
(the host ID of 63 is the broadcast address). We’ll use the host ID 36 in this example.

bsdi % date ; telnet 140.252.13.36 ; date
Sat Jan 30 06:46:33 MST 1993 telnet to an address this time,
Trying 140.252.13.36... not a hostname
telnet: Unable to connect to remote host: 76 seconds after previous date
Connection timed out output
Sat Jan 30 06:47:49 MST 1993
bsdi % arp -a
check the ARP cache
? (140.252.13.36) at (incomplete)

Figure 4.5 shows the tcpdump output.

64

1 0.0 arp who has 140.252.13.36 tell bsdi
2 5.509069 (5.5091) arp who has 140.252.13.36 tell bsdi
3 29.509745 (24.0007) arp who has 140.252.13.36 tell bsdi

Figure 4.5 ARP requests to a nonexistent host.

This time we didn’t specify the -e option since we already know that the ARP requests are broadcast.

What’s interesting here is to see the frequency of the ARP requests: 5.5 seconds after the first request,
then again 24 seconds later. (We examine TCP’s timeout and retransmission algorithms in more detail
in Chapter 21.) The total time shown in the tcpdump output is 29.5 seconds. But the output from the
date commands before and after the telnet command shows that the connection request from the
Telnet client appears to have given up after about 75 seconds. Indeed, we’ll see later that most BSD
implementations set a limit of 75 seconds for a TCP connection request to complete.

In Chapter 18 when we see the sequence of TCP segments that is sent to establish the connection,
we’ll see that these ARP requests correspond one-to-one with the initial TCP SYN (synchronize)
segment that TCP is trying to send.

Note that on the wire we never see the TCP segments. All we can see are the ARP requests. Until an
ARP reply comes back, the TCP segments can’t be sent, since the destination hardware address isn’t
known. If we ran tcpdump in a filtering mode, looking only for TCP data, there would have been no
output at all.

ARP Cache Timeout
A timeout is normally provided for entries in the ARP cache. (In Section 4.8 we’ll see that the arp
command allows an entry to be placed into the cache by the administrator that will never time out.)
Berkeley-derived implementations normally have a timeout of 20 minutes for a completed entry and 3
minutes for an incomplete entry (We saw an incomplete entry in our previous example where we
forced an ARP to a nonexistent host on the Ethernet.) These implementations normally restart the
20-minute timeout for an entry each time the entry is used.

The Host Requirements RFC says that this timeout should occur even if the entry is in use, but most Berkeley-derived
implementations do not do this-they restart the timeout each time the entry is referenced.

4.6 Proxy ARP
Proxy ARP lets a router answer ARP requests on one of its networks for a host on another of its
networks. This fools the sender of the ARP request into thinking that the router is the destination host,
when in fact the destination host is "on the other side" of the router. The router is acting as a proxy
agent for the destination host, relaying packets to it from other hosts.

An example is the best way to describe proxy ARP. In Figure 3.10 we showed that the system sun
was connected to two Ethernets. But we also noted that this wasn’t really true, if you compare that
figure with the one on the inside front cover. There is in fact a router between sun and the subnet
140.252.1, and this router performs proxy ARP to make it appear as though sun is actually on the
subnet 140.252.1. Figure 4.6 shows the arrangement, with a Telebit NetBlazer, named netb, between
the subnet and the host sun.

65

Figure 4.6 Example of proxy ARP.

When some other host on the subnet 140.252.1 (say, gemini) has an IP datagram to send to sun at
address 140.252.1.29, gemini compares the network ID (140.252) and subnet ID (1) and since they
are equal, issues an ARP request on the top Ethernet in Figure 4.6 for IP address 140.252.1.29. The
router netb recognizes this IP address as one belonging to one of its dialup hosts, and responds with
the hardware address of its Ethernet interface on the cable 140.252.1. The host gemini sends the IP
datagram to netb across the Ethernet, and netb forwards the datagram to sun across the dialup
SLIP link. This makes it transparent to all the hosts on the 140.252.1 subnet that host sun is really
configured "behind" the router netb.

If we execute the arp command on the host gemini, after communicating with the host sun, we see
that both IP addresses on the 140.252.1 subnet, netb and sun, map to the same hardware address.
This is often a clue that proxy ARP is being used.

many lines for other hosts on the 140.252.1
gemini % arp -a
subnet
netb (140.252.1.183) at
0:80:ad:3:6a:80
sun (140.252.1.29) at
0:80:ad:3:6a:80

Another detail in Figure 4.6 that we need to explain is the apparent lack of an IP address at the bottom
of the router netb (the SLIP link). That is, why don’t both ends of the dialup SLIP link have an IP
address, as do both ends of the hardwired SLIP link between bsdi and slip? We noted in Section
3.8 that the destination address of the dialup SLIP link, as shown by the ifconfig command, was
140.252.1.183. The Net-Blazer doesn’t require an IP address for its end of each dialup SLIP link.
(Doing so would use up more IP addresses.) Instead, it determines which dialup host is sending it
packets by which serial interface the packet arrives on, so there’s no need for each dialup host to use a
unique IP address for its link to the router. All the dialup hosts use 140.252.1.183 as the destination
address for their SLIP link.

Proxy ARP handles the delivery of datagrams to the router sun, but how are the other hosts on the
subnet 140.252.13 handled? Routing must be used to direct datagrams to the other hosts. Specifically,
routing table entries must be made somewhere on the 140.252 network that point all datagrams
destined to either the subnet 140.252.13, or the specific hosts on that subnet, to the router netb. This
router then knows how to get the datagrams to their final destination, by sending them through the
router sun.

Proxy ARP is also called promiscuous ARP or the ARP hack. These names are from another use of
proxy ARP: to hide two physical networks from each other, with a router between the two. In this case
both physical networks can use the same network ID as long as the router in the middle is configured
as a proxy ARP agent to respond to ARP requests on one network for a host on the other network. This
technique has been used in the past to "hide" a group of hosts with older implementations of TCP/IP

66

on a separate physical cable. Two common reasons for separating these older hosts are their inability
to handle subnetting and their use of the older broadcasting address (a host ID of all zero bits, instead
of the current standard of a host ID with all one bits).

4.7 Gratuitous ARP
Another feature of ARP that we can watch is called gratuitous ARP. It occurs when a host sends an
ARP request looking for its own IP address. This is usually done when the interface is configured at
bootstrap time.

In our internet, if we bootstrap the host bsdi and run tcpdump on the host sun, we see the packet
shown in Figure 4.7.

0:0:c0:6f:2d:40 ff:ff:ff:ff:ff:ff arp 60:
1 0.0
arp who has 140.252.13.35 tell 140.252.13.35

Figure 4.7 Example of gratuitous ARP.

(We specified the -n flag for tcpdump to print numeric dotted-decimal addresses, instead of
hostnames.) In terms of the fields in the ARP request, the sender’s protocol address and the target’s
protocol address are identical: 140.252.13.35 for host bsdi. Also, the source address in the Ethernet
header, 0:0:c0:6f:2d:40 as shown by tcpdump, equals the sender’s hardware address (from
Figure 4.4). Gratuitous ARP provides two features.

1. It lets a host determine if another host is already configured with the same IP address. The host
bsdi is not expecting a reply to this request. But if a reply is received, the error message
"duplicate IP address sent from Ethernet address: a:b:c:d:e:f" is logged on the console. This is a
warning to the system administrator that one of the systems is misconfigured.

2. If the host sending the gratuitous ARP has just changed its hardware address (perhaps the host
was shut down, the interface card replaced, and then the host was rebooted), this packet causes
any other host on the cable that has an entry in its cache for the old hardware address to update its
ARP cache entry accordingly. A little known fact of the ARP protocol [Plummer 1982] is that if a
host receives an ARP request from an IP address that is already in the receiver’s cache, then that
cache entry is updated with the sender’s hardware address (e.g., Ethernet address) from the ARP
request. This is done for any ARP request received by the host. (Recall that ARP requests are
broadcast, so this is done by all hosts on the network each time an ARP request is sent.)

[Bhide, Einozahy, and Morgan 1991] describe an application that can use this feature of ARP to
allow a backup file server to take over from a failed server by issuing a gratuitous ARP request
with the backup’s hardware address and the failed server’s IP address. This causes all packets
destined for the failed server to be sent to the backup instead, without the client applications
being aware that the original server has failed.

Unfortunately the authors then decided against this approach, since it depends on the correct
implementation of ARP on all types of clients. They obviously encountered client
implementations that did not implement ARP according to its specification.

Monitoring all the systems on the author’s subnet shows that SunOS 4.1.3 and 4.4BSD both issue gratuitous ARPs
when bootstrapping, but SVR4 does not.

67

4.8 arp Command
We’ve used this command with the -a flag to display all the entries in the ARP cache. Other options
are provided.

The superuser can specify the -d option to delete an entry from the ARP cache. (This was used before
running a few of the examples, to let us see the ARP exchange.)

Entries can also be added using the -soption. It requires a hostname and an Ethernet address: the IP
address corresponding to the hostname, and the Ethernet address are added to the cache. This entry is
made permanent (i.e., it won’t time out from the cache) unless the keyword temp appears at the end
of the command line.

The keyword pub at the end of a command line with the -s option causes the system to act as an ARP
agent for that host. The system will answer ARP requests for the IP address corresponding to the
hostname, replying with the specified Ethernet address.If the advertised address is the system’s own,
then this system is acting as a proxy ARP agent for the specified hostname.

4.9 Summary
ARP is a basic protocol in almost every TCP/IP implementation, but it normally does its work without
the application or the system administrator being aware. The ARP cache is fundamental to its
operation, and we’ve used the arp command to examine and manipulate the cache. Each entry in the
cache has a timer that is used to remove both incomplete and completed entries. "The arp command
displays and modifies entries in the ARP cache.

We followed through the normal operation of ARP along with specialized versions: proxy ARP (when
a router answers ARP requests for hosts accessible on another of the router’s interfaces) and gratuitous
ARP (sending an ARP request for your own IP address, normally when bootstrapping).

Exercises
4.1 In the commands we typed to generate the output shown in Figure 4.4, what would happen if, after
verifying that the local ARP cache was empty, we type the command

bsdi % rsh svr4 arp -a

to verify that the ARP cache is also empty on the destination host? (This command causes the arp
-a command to be executed on the host svr4.)

4.2 Describe a test to determine if a given host handles a received gratuitous ARP request correctly

4.3 Step 7 in Section 4.2 can take a while (milliseconds) because a packet is sent and ARP then waits
for the response. How do you think ARP handles multiple datagrams that arrive from IP for the same
destination address during this period?

4.4 At the end of Section 4.5 we mentioned that the Host Requirements RFC and Berkeley-derived
implementations differ in their handling of the timeout of an active ARP entry. What happens if we’re
on a Berkeley-derived client and keep trying to contact a server host that’s been taken down to replace
its Ethernet board? Does this change if the server issues a gratuitous ARP when it bootstraps?

68

RARP: Reverse Address Resolution Protocol
5.1 Introduction
When a system with a local disk is bootstrapped it normally obtains its IP address from a configuration
file that’s read from a disk file. But a system without a disk, such as an X terminal or a diskless
workstation, needs some other way to obtain its IP address.

Each system on a network has a unique hardware address, assigned by the manufacturer of the
network interface. The principle of RARP is for the diskless system to read its unique hardware
address from the interface card and send an RARP request (a broadcast frame on the network) asking
for someone to reply with the diskless system’s IP address (in an RARP reply).

While the concept is simple, the implementation is often harder than ARP for reasons described later
in this chapter. The official specification of RARP is RFC 903 [Finlayson et al. 1984].

5.2 RARP Packet Format
The format of an RARP packet is almost identical to an ARP packet (Figure 4.3). The only differences
are that the frame type is 0x8035 for an RARP request or reply, and the op field has a value of 3 for
an RARP request and 4 for an RARP reply.

As with ARP, the RARP request is broadcast and the RARP reply is normally unicast.

5.3 RARP Examples
In our internet we can force the host sun to bootstrap from the network, instead of its local disk. If we
run an RARP server and tcpdump on the host bsdi we get the output shown in Figure 5.1. We use the
-e flag to have tcpdump print the hardware addresses:

8:0:20:3:f6:42 ff:ff:ff:ff:ff:ff
1 0.0 rarp 60: rarp who-is 8:0:20:3:f6:42 tell
8:0:20:3:f6:42
0.13 0:0:c0:6f:2d:40 8:0:20:3:f6:42 rarp 42:
2
(0.13) rarp reply 8:0:20:3:f6:42 at sun
0.14 8:0:20:3:f6:42 0:0:c0:6f:2d:40 ip 65:
3
(0.01) >sun.26999 > bsdi.tftp: 23 RRQ "8CFCOD21.SUN4C"

Figure 5.1 RARP request and reply.

The RARP request is broadcast (line 1) and the RARP reply on line 2 is unicast. The output on line 2,
at sun, means the RARP reply contains the IP address for the host sun (140.252.13.33).

On line 3 we see that once sun receives its IP address, it issues a TFTP read-request (RRQ) for the file
8CFCOD21.SUN4C. (TFTP is the Trivial File Transfer Protocol. We describe it in more detail in
Chapter 15.) The eight hexadecimal digits in the filename are the hex representation of the IP address
140.252.13.33 for the host sun. This is the IP address that was returned in the RARP reply. The
remainder of the filename, SUN4C, indicates the type of system being bootstrapped.

69

tcpdump says that line 3 is an IP datagram of length 65, and not a UDP datagram (which it really is),
because we are running tcpdump with the -eflag, to see the hardware-level addresses. Another point
to notice in Figure 5.1 is that the length of the Ethernet frame on line 2 appears to be shorter than the
minimum (which we said was 60 bytes in Section 4.5.) The reason is that we are running tcpdump on
the system that is sending this Ethernet frame (bsdi). The application, rarpd, writes 42 bytes to the
BSD Packet Filter device (14 bytes for the Ethernet header and 28 bytes for the RARP reply) and this
is what tcpdump receives a copy of. But the Ethernet device driver pads this short frame to the
minimum size for transmission (60). Had we been running tcpdump on another system, the length
would have been 60.

We can see in this example that when this diskless system receives its IP address in an RARP reply, it
issues a TFTP request to read a bootstrap image. At this point we won’t go into additional detail about
how diskless systems bootstrap themselves. (Chapter 16 describes the bootstrap sequence of a diskless
X terminal using RARP, BOOTP, and TFTP.)

Figure 5.2 shows the resulting packets if there is no RARP server on the network. The destination
address of each packet is the Ethernet broadcast address. The Ethernet address followingwho-is is
the target hardware address, and the Ethernet address following tell is the sender’s hardware
address.

70

8:0:20:3:f6:42 ff:ff:ff:ff:ff:ff rarp 60:
1 0.0 rarp who-is 8:0:20:3:f6:42 tell
8:0:20:3;f6:42
2 6.55 ( 6.55) rarp who-is 8:0:20:3:f6:42 tell
8:0:20:3:f6:42
3 15.52 ( 8.97) rarp who-is 8:0:20:3:f6:42 tell
8:0:20:3:f6:42
8:0:20:3:f6:42 ff:ff:ff;ff:ff:ff rarp 60:
4 29.32 (13.80) rarp who-is 8:0:20:3:f6:42 tell
8:0:20:3:f6:42
5 52.78 (23.46) rarp who-is 8:0:20:3:f6:42 tell
8:0:20:3:f6:42
6 95.58 (42.80) rarp who-is 8:0:20:3:f6:42 tell
8:0;20:3:f6:42
100.92 (
7 rarp who-is 8:0:20:3;f6:42 tell
5.34)
8:0:20:3:f6:42
107.47 (
8 rarp who-is 8:0:20:3:f6:42 tell
6.55)
8:0:20:3:f6:42
116.44 (
8.97)
8:0:20:3:f6:42
130.24
(13.80)
8:0:20:3:f6:42
153.70
(23.46)
8:0:20:3:f6:42
196.49
12 rarp who-is 8:0;20:3:f6:42 tell
(42.79)
8:0:20:3:f6:42

Figure 5.2 RARP requests with no RARP server on the network.

Note the frequency of the retransmissions. The first retransmission occurs after 6.55 seconds and then
increases to 42.80 seconds, then goes down to 5.34 seconds, then 6.55, and then works its way back to
42.79 seconds. This continues indefinitely. If we calculate the differences between each timeout
interval we see a doubling effect: from 5.34 to 6.55 is 1.21 seconds, from 6.55 to 8.97 is 2.42 seconds,

71

from 8.97 to 13.80 is 4.83 seconds, and so on. When the timeout interval reaches some limit (greater
than 42.80 seconds) it’s reset to 5.34 seconds.

Increasing the timeout value like this is a better approach than using the same value each time. In
Figure 6.8 we’ll see one wrong way to perform timeout and retransmission, and in Chapter 21 we’ll
see TCP’s method.

5.4 RARP Server Design
While the concept of RARP is simple, the design of an RARP server is system dependent and
complex. Conversely, providing an ARP server is simple, and is normally part of the TCP/IP
implementation in the kernel. Since the kernel knows its IP addresses and hardware addresses, when it
receives an ARP request for one of its IP addresses, it just replies with the corresponding hardware
address.

RARP Servers as User Processes
The complication with an RARP server is that the server normally provides the mapping from a
hardware address to an IP address for many hosts (all the diskless systems on the network). This
mapping is contained in a disk file (normally /etc/ethers on Unix systems). Since kernels
normally don’t read and parse disk files, the function of an RARP server is provided as a user process,
not as part of the kernel’s TCP/IP implementation.

To further complicate matters, RARP requests are transmitted as Ethernet frames with a specific
Ethernet frame type field (0x8035 from Figure 2.1.) This means an RARP server must have some
way of sending and receiving Ethernet frames of this type. In Appendix A we describe how the BSD
Packet Filter, Sun’s Network Interface Tap, and the SVR4 Data Link Provider Interface can be used to
receive these frames. Since the sending and receiving of these frames is system dependent, the
implementation of an RARP server is tied to the system.

Multiple RARP Servers per Network
Another complication is that RARP requests are sent as hardware-level broadcasts, as shown in Figure
5.2. This means they are not forwarded by routers. To allow diskless systems to bootstrap even when
the RARP server host is down, multiple RARP servers are normally provided on a single network
(e.g., a single cable).

As the number of servers increases (to provide redundancy), the network traffic increases, since every
server sends an RARP reply for every RARP request. The diskless system that sent the RARP request
normally uses the first RARP reply that it receives. (We never had this problem with ARP, because
only a single host sends an ARP reply.) Furthermore, there is a chance that each RARP server can try
to respond at about the same time, increasing the probability of collisions on an Ethernet.

5.5 Summary
RARP is used by many diskless systems to obtain their IP address when bootstrapped. The RARP
packet format is nearly identical to the ARP packet. An RARP request is broadcast, identifying the
sender’s hardware address, asking for anyone to respond with the sender’s IP address. The reply is
normally unicast.

72

Problems with RARP include its use of a link-layer broadcast, preventing most routers from
forwarding an RARP request, and the minimal information returned: just the system’s IP address. In
Chapter 16 we’ll see that BOOTP returns more information for the diskless system that is
bootstrapping: its IP address, the name of a host to bootstrap from, and so on.

While the RARP concept is simple, the implementation of an RARP server is system dependent.
Hence not all TCP/IP implementations provide an RARP server.

Exercises
5.1 Is a separate frame type field required for RARP? Could the same value be used for ARP and
RARP 0x0806?

5.2 With multiple RARP servers on a network, how can they prevent their responses from colliding
with each on the network?

73

ICMP: Internet Control Message Protocol
6.1 Introduction
ICMP is often considered part of the IP layer. It communicates error messages and other conditions
that require attention. ICMP messages are usually acted on by either the IP layer or the higher layer
protocol (TCP or UDP). Some ICMP messages cause errors to be returned to user processes.

ICMP messages are transmitted within IP datagrams, as shown in Figure 6.1.

Figure 6.1 ICMP messages encapsulated within an IP datagram.

RFC 792 [Postel 1981b] contains the official specification of ICMP.

Figure 6.2 shows the format of an ICMP message. The first 4 bytes have the same format for all
messages, but the remainder differs from one message to the next. We’ll show the exact format of each
message when we describe it.

There are 15 different values for the type field, which identify the particular ICMP message. Some
types of ICMP messages then use different values of the code field to further specify the condition.

The checksum field covers the entire ICMP message. The algorithm used is the same as we described
for the IP header checksum in Section 3.2. The ICMP checksum is required.

Figure 6.2 ICMP message.

In this chapter we talk about ICMP messages in general and a few in detail: address mask request and
reply, timestamp request and reply, and port unreachable. We discuss the echo request and reply
messages in detail with the Ping program in Chapter 7, and we discuss the ICMP messages dealing
with IP routing in Chapter 9.

6.2 ICMP Message Types
Figure 6.3 lists the different ICMP message types, as determined by the type field and code field in the
ICMP message.

74

The final two columns in this figure specify whether the ICMP message is a query message or an error
message. We need to make this distinction because ICMP error messages are sometimes handled
specially. For example, an ICMP error message is never generated in response to an ICMP error
message. (If this were not the rule, we could end up with scenarios where an error generates an error,
which generates an error, and so on, indefinitely)

When an ICMP error message is sent, the message always contains the IP header and the first 8 bytes
of the IP datagram that caused the ICMP error to be generated. This lets the receiving ICMP module
associate the message with one particular protocol (TCP or UDP from the protocol field in the IP
header) and one particular user process (from the TCP or UDP port numbers that are in the TCP or
UDP header contained in the first 8 bytes of the IP datagram). We’ll show an example of this in
Section 6.5. An ICMP error message is never generated in response to

1. An ICMP error message. (An ICMP error message may, however, be generated in response to an
ICMP query message.)
2. A datagram destined to an IP broadcast address (Figure 3.9) or an IP multicast address (a class D
address, Figure 1.5).
3. A datagram sent as a link-layer broadcast.
4. A fragment other than the first. (We describe fragmentation in Section 11.5.)
5. A datagram whose source address does not define a single host. This means the source address
cannot be a zero address, a loopback address, a broadcast address, or a multicast address.

75

type code Description Query Error
0 0 echo reply (Ping reply. Chapter 7) *
destination unreachable:
0 *
network unreachable (Section 9.3)
1 *
host unreachable (Section 9.3)
2 *
protocol unreachable
3 *
port unreachable (Section 6.5)
4 *
fragmentation needed but don’t-fragment bit set (Section 11.6)
5 *
source route failed (Section 8.5)
6 *
destination network unknown
7 *
3 destination host unknown
8 *
source host isolated (obsolete)
9 *
destination network administratively prohibited
10 *
destination host administratively prohibited
11 *
network unreachable for TOS (Section 9.3)
12 *
host unreachable for TOS (Section 9.3)
13 *
communication administratively prohibited by filtering
14 *
host precedence violation
15 *
precedence cutoff in effect
4 0 source quench (elementary flow control. Section 11.11) *
redirect (Section 9.5):
0 *
redirect for network
1 *
5 redirect for host
2 *
redirect for type-of-service and network
3 *
redirect for type-of-service and host
8 0 echo request (Ping request. Chapter 7) *
9 0 router advertisement (Section 9.6) *
10 0 router solicitation (Section 9.6) *
time exceeded:
0 *
11 time-to-live equals 0 during transit (Traceroute, Chapter 8)
1 *
time-to-live equals 0 during reassembly (Section 11.5)
parameter problem:
0 *
12 IP header bad (catchall error)
1 *
required option missing
13 0 timestamp request (Section 6.4) *
14 0 timestamp reply (Section 6.4) *
15 0 information request (obsolete) *
16 0 information reply (obsolete) *
17 0 address mask request (Section 6.3) *
18 0 address mask reply (Section 6.3) *

76

Figure 6.3 ICMP message types.

These rules are meant to prevent the broadcast storms that have occurred in the past when ICMP
errors were sent in response to broadcast packets.

6.3 ICMP Address Mask Request and Reply
The ICMP address mask request is intended for a diskless system to obtain its subnet mask (Section
3.5) at bootstrap time. The requesting system broadcasts its ICMP request. (This is similar to a diskless
system using RARP to obtain its IP address at bootstrap time.) An alternative method for a diskless
system to obtain its subnet mask is the BOOTP protocol, which we describe in Chapter 16. Figure 6.4
shows the format of the ICMP address mask request and reply messages.

Figure 6.4 ICMP address mask request and reply messages.

The identifier and sequence number fields in the ICMP message can be set to anything the sender
chooses, and these values are returned in the reply This allows the sender to match replies with
requests.

We can write a simple program (named icmpaddrmask) that issues an ICMP address mask request
and prints all replies. Since normal usage is to send the request to the broadcast address, that’s what
we’ll do. The destination address (140.252.13.63) is the broadcast address for the subnet
140.252.13.32 (Figure 3.12).

sun % icmpaddnnask 140.252.13.63
from ourself
received mask = ffffffeO, from 140.252.13.33
from bsdi
received mask = ffffffeO, from 140.252.13.35
from svr4
received mask = ffff0000, from 140.252.13.34

The first thing we note in this output is that the returned value from svr4 is wrong. It appears that
SVR4 is returning the general class B address mask, assuming no subnets, even though the interface
on svr4 has been configured with the correct subnet mask:

svr4 % ifconfig emd0
emd0: flags=23<UP, BROADCAST ,NOTRAILERS>

There is a bug in the SVR4 handling of the ICMP address mask request.

We’ll watch this exchange on the host bsdi using tcpdump. The output is shown in Figure 6.5. We
specify the -e option to see the hardware addresses.

77

8:0:20:3:f6:42 ff;ff:ff:ff:ff:ff ip 60:
1 0.0
sun > 140.252.13.63: icmp: address mask request
0:0:c0:6f:2d:40 ff;ff:ff:ff:ff:ff ip 46:
2 0.00 (0.00)
bsdi > sun: icmp: address mask is 0xffffffe0
0:0:c0:c2:9b:26 8:0:20:3:f6:42 ip 60:
3 0.01 (0.01)
svr4 > sun: icmp: address mask is 0xffff0000

Figure 6.5 ICMP address mask request sent to broadcast address.

Note that the sending host, sun, receives an ICMP reply (the output line with the comment from
ourself shown earlier), even though nothing is seen on the wire. This is a general characteristic of
broadcasting: the sending host receives a copy of the broadcast packet through some internal loopback
mechanism. Since by definition the term "broadcast" means all the hosts on the local network, it
should include the sending host. (Referring to Figure 2.4 what is happening is that when the Ethernet
driver recognizes that the destination address is the broadcast address, the packet is sent onto the
network and a copy is made and passed to the loopback interface.)

Next, bsdi broadcasts the reply, while svr4 sends the reply only to the requestor. Normally the
reply should be unicast unless the source IP address of the request is 0.0.0.0, which it isn’t in this
example. Therefore, sending the reply to the broadcast address is a BSD/386 bug.

The Host Requirements RFC says that a system must not send an address mask reply unless it is an
authoritative agent for address masks. (To be an authoritative agent it must be specifically configured
to send these replies. See Appendix E.) As we can see from this example, however, most host
implementations send a reply if they get a request. Some hosts even send the wrong reply!

The final point is shown by the following example. We send an address mask request to our own IP
address and to the loopback address:

sun % icmpaddrmask sun
received mask = ff000000, from 140.252.13.33
sun % icmpaddrmask localhost
received mask = ff000000, from 127.0.0.1

In both cases the returned address mask corresponds to the loopback address, the class A address
127.0.0.1. Again, referring to Figure 2.4 we see that IP datagrams sent to the host’s own IP address
(140.252.13.33 in this example) are actually sent to the loop-back interface. The ICMP address mask
reply must correspond to the subnet mask of the interface on which the request was received (since a
multihomed host can have different subnet masks for each interface), and in both cases the request is
received from the loopback interface.

6.4 ICMP Timestamp Request and Reply
The ICMP timestamp request allows a system to query another for the current time. The recommended
value to be returned is the number of milliseconds since midnight, Coordinated Universal Time
(UTC). (Older manuals refer to UTC as Greenwich Mean Time.) The nice feature of this ICMP
message is that it provides millisecond resolution, whereas some other methods for obtaining the time
from another host (such as the rdate command provided by some Unix systems) provide a resolution

78

of seconds. The drawback is that only the time since midnight is returned-the caller must know the
date from some other means. Figure 6.6 shows the format of the ICMP timestamp request and reply
messages.

Figure 6.6 ICMP timestamp request and reply messages.

The requestor fills in the originate timestamp and sends the request. The replying system fills in the
receive timestamp when it receives the request, and the transmit time-stamp when it sends the reply. In
actuality, however, most implementations set the latter two fields to the same value. (The reason for
providing the three fields is to let the sender compute the time for the request to be sent, and separately
compute the time for the reply to be sent.)

We can write a simple program (named icmptime) that sends an ICMP timestamp request to a host
and prints the returned reply. We try it first on our small internet:

sun % icmptime bsdi
orig = 83573336, recv = 83573330, xmit = 83573330, rtt = 2 ms
difference = -6 ms
sun % icmptime bsdi
difference = -7 ms

The program prints the three timestamps in the ICMP message: the originate (orig), receive (recv),
and transmit (xmit) timestamps. As we can see in this and the following examples, all the hosts set
the receive and transmit timestamps to the same value.

We also calculate the round-trip time (rtt), which is the time the reply is received minus the time the
request was sent. The difference is the received timestamp minus the originate timestamp. Figure
6.7 shows the relationship between these values.

Figure 6.7 Relationship between values printed by our icmptime program.

If we believe the RTT and assume that one-half of the RTT is for the request, and the other half for the
reply, then the sender’s clock needs to be adjusted by difference minus one-half the RTT, to have
the same time as the host being queried. In the preceding example, the clock on bsdi was 7 and 8 ms
behind the clock on sun.

79

Since the timestamp values are the number of milliseconds past midnight, UTC, they should always be
less than 86,400,000 (24 x 60 x 60 x 1000). These examples were run just before 4:00 P.M. in a time
zone that is 7 hours behind UTC, so the values being greater than 82,800,000 (2300 hours) makes
sense.

If we run this program several times to the host bsdi we see that the final digit in the receive and
transmit timestamp is always 0. This is because the software release (Version 0.9.4) only provides a
10-ms clock. (We describe this in Appendix B.)

If we run the program twice to the host svr4 we see that the low-order three digits of the SVR4
timestamp are always 0:

sun % icmptime svr4
difference = -210 ms
sun % icmptime svr4

For some reason SVR4 doesn’t provide any millisecond resolution using the ICMP time-stamp. This
imprecision makes the calculated differences useless for subsecond adjustments.

If we try two other hosts on the 140.252.1 subnet, the results show that one clock differs from sun’s by
3.7 seconds, and the other by nearly 75 seconds:

sun % icmptime gemini
sun % icmptime aix

Another interesting example is to the router gateway (a Cisco router). It shows that when a system
returns a nonstandard timestamp value (something other than milliseconds past midnight, UTC), it is
supposed to turn on the high-order bit of the 32-bit timestamp. Our program detects this, and prints the
receive and transmit timestamps in angle brackets (after turning off the high-order bit). Also, we can’t
calculate the difference between the originate and receive timestamps, since they’re not the same units.

sun % icmptime gateway
orig = 83620811, recv = <4871036>, xmit = <4871036>, rtt = 220 ms
sun % icmptime gateway
orig = 83641007, recv = <4891232>, xmit = <4891232>, rtt = 213 ms

If we run our program to this host a few times it becomes obvious that the values do contain
millisecond resolution and do count the number of milliseconds past some starting point, but the
starting point is not midnight, UTC. (It could be a counter that’s incremented every millisecond since
the router was bootstrapped, for example.)

80

As a final example we’ll compare sun’s clock with a system whose clock is known to be accurate-an
NTP stratum 1 server. (We say more about NTP, the Network Time Protocol, below.)

sun % icmptime clock.llnl.gov
difference = 128 ms
sun % icmptime clock.llnl.gov
difference = 134 ms

If we calculate the difference minus one-half the RTT, this output indicates that the clock on sun is
between 38.5 and 51.5 ms fast.

Alternatives
There are other ways to obtain the time and date.

1. We described the daytime service and time service in Section 1.12. The former returns the current
time and date in a human readable form, a line of ASCII characters. We can test this service using
the telnet command:

sun % telnet bsdi daytime
Trying 140.252.13.35 ...
first three lines output are from the Telnet
Connected to bsdi.
client
Escape character is "^]"
here’s the daytime service output
Wed Feb 3 16:38:33 1993
this is also from the Telnet client
Connection closed by foreign
host.

The time server, on the other hand, returns a 32-bit binary value with the number of seconds since
midnight January 1, 1900, UTC. While this provides the date, the time value is in units of a
second. (The rdate command that we mentioned earlier uses the TCP time service.)

2. Serious timekeepers use the Network Time Protocol (NTP) described in RFC 1305 [Mills 1992].
This protocol uses sophisticated techniques to maintain the clocks for a group of systems on a
LAN or WAN to within millisecond accuracy. Anyone interested in precise timekeeping on
computers should read this RFC.

3. The Open Software Foundation’s (OSF) Distributed Computing Environment (DCE) defines a
Distributed Time Service (DTS) that also provides clock synchronization between computers.
[Rosenberg, Kenney, and Fisher 1992] provide additional details on this service.

4. Berkeley Unix systems provide the daemon timed(8) to synchronize the clocks of systems on a
local area network. Unlike NTP and DTS, timed does not work across wide area networks.

81

6.5 ICMP Port Unreachable Error
The last two sections looked at ICMP query messages-the address mask and time-stamp queries and
replies. We’ll now examine an ICMP error message, the port unreachable message, a subcode of the
ICMP destination unreachable message, to see the additional information returned in an ICMP error
message. We’ll watch this using UDP (Chapter 11).

One rule of UDP is that if it receives a UDP datagram and the destination port does not correspond to a
port that some process has in use, UDP responds with an ICMP port unreachable. We can force a port
unreachable using the TFTP client. (We describe TFTP in Chapter 15.)

The well-known UDP port for the TFTP server to be reading from is 69. But most TFTP client
programs allow us to specify a different port using the connect command. We use this to specify a port
of 8888:

bsdi % tftp
tftp> connect svr4 8888 specify the hostname and port number
tftp> get temp.foo try to fetch a file
Transfer timed out. about 15 seconds later
tftp> quit

The connect command saves the name of the host to contact and the port number on that host, for
when we later issue the get command. After typing the get command a UDP datagram is sent to port
8888 on host svr4. Figure 6.8 shows the tcpdump output for the exchange of packets that takes
place.

Before the UDP datagram can be sent to svr4 an ARP request is sent to determine its hardware address
(line 1). The ARP reply (line 2) is returned and then the UDP datagram is sent (line 3). (We have left
the ARP request-reply in this tcpdump output to remind us that this exchange may be required before
the first IP datagram is sent from one host to the other. In future output we’ll delete this exchange if
it’s not relevant to the topic being discussed.)

1 0.0 arp who-has svr4 tell bsdi
2 0.002050 (0.0020) arp reply svr4 is-at 0:0:c0:c2:9b:26
3 0.002723 (0.0007) bsdi.2924 > svr4.8888: udp 20
4 0.006399 (0.0037) svr4 > bsdi: icmp: svr4 udp port 8888
5 5.000776 (4.9944) bsdi. 2924 > svr4.8888: udp 20
7 10.000887 (4.9966) bsdi. 2924 > svr4.8888: udp 20
9 15.001014 (4.9966) bsdi.2924 > svr4.8888: udp 20
11 20.001177 (4.9966) bsdi.2924 > svr4.8888: udp 20

82

Figure 6.8 ICMP port unreachable generated by TFTP request.

An ICMP port unreachable is immediately returned (line 4). But the TFTP client appears to ignore the
ICMP message, sending another UDP datagram about 5 seconds later (line 5). This continues three
more times before the client gives up.

Notice that the ICMP messages are exchanged between hosts, without a port number designation,
while each 20-byte UDP datagram is from a specific port (2924) and to a specific port (8888).

The number 20 at the end of each UDP line is the length of the data in the UDP datagram. In this
example 20 is the sum of the TFTP’s 2-byte opcode, the 9-byte null terminated name temp.foo, and
the 9-byte null terminated string netascii. (See Figure 15.1 for the details of the TFTP packet
layout.)

If we run this same example using the -e option of tcpdump we see the exact length of each ICMP
port unreachable message that’s returned to the sender. This length is 70 bytes, and is allocated as

Figure 6.9 ICMP message returned for our "UDP port unreachable" example.

One rule of ICMP is that the ICMP error messages (see the final column of Figure 6.3) must include
the IP header (including any options) of the datagram that generated the error along with at least the
first 8 bytes that followed this IP header. In our example, the first 8 bytes following the IP header
contain the UDP header (Figure 11.2).

The important fact is that contained in the UDP header are the source and destination port numbers. It
is this destination port number (8888) that caused the ICMP port unreachable to be generated. The
source port number (2924) can be used by the system receiving the ICMP error to associate the error
with a particular user process (the TFTP client in this example).

One reason the IP header of the datagram that caused the error is sent back is because in this IP header
is the protocol field that lets ICMP know how to interpret the 8 bytes that follow (the UDP header in
this example). When we look at the TCP header (Figure 17.2) we’ll see that the source and destination
port numbers are contained in the first 8 bytes of the TCP header. The general format of the ICMP
unreachable messages is shown in Figure 6.10.

Figure 6.10 ICMP unreachable message.

83

In Figure 6.3 we noted that there are 16 different ICMP unreachable messages, codes 0 through 15.
The ICMP port unreachable is code 3. Also, although Figure 6.10 indicates that the second 32-bit word
in the ICMP message must be 0, the Path MTU Discovery mechanism (Section 2.9) allows a router to
place the MTU of the outgoing interface in the low-order 16 bits of this 32-bit value, when code equals
4 ("fragmentation needed but the don’t fragment bit is set"). We show an example of this error in
Section 11.6.

Although the rules of ICMP allow a system to return more than the first 8 bytes of the data portion of the IP datagram that
caused the ICMP error, most Berkeley-derived implementations return exactly 8 bytes. The Solaris 2.2
ip_icmp_return_data_bytes option returns the first 64 bytes of data by default (Section E.4).

tcpdump Time Line
Throughout the text we’ll also display the tcpdump output in a time line diagram as shown in Figure
6.11.

Figure 6.11 Time line of TFTP request to an invalid port.

Time increases down the page and the labels on the far left of the figure are the same time values as in
our tcpdump output (Figure 6.8). The labels at the top are the hostnames and port numbers for each
side of the time line. Be aware that the y-axis down the page is not exactly proportional to the time
value. When there is a significant time lag, as between each 5-second retransmission in this example,
we’ll designate that with a squiggle on both sides of the time line. When UDP or TCP data is being
transmitted, we show that packet with a thicker line.

Why does the TFTP client keep retransmitting its request when the ICMP messages are being
returned? An element of network programming is occurring in which BSD systems don’t notify user
processes using UDP of ICMP messages that are received for that socket unless the process has issued
a connect on that socket. The standard BSD TFTP client does not issue the connect, so it never
receives the ICMP error notification.

Another point to notice here is the poor retransmission timeout algorithm used by this TFTP client. It
just assumes that 5 seconds is adequate and retransmits every 5 seconds, for a total of 25 seconds.
We’ll see later that TCP has a much better algorithm.

This old-fashioned timeout and retransmission algorithm used by the TFTP client is forbidden by the Host Requirements
RFC. Nevertheless, all three systems on the author’s subnet, and Solaris 2.2 still use it. AIX 3.2.2 applies an exponential
backoff to its timeout, sending packets at 0, 5, 15, and 35 seconds, which is the recommended way We talk much more
about timeouts in Chapter 21.

Finally note that the ICMP messages are returned about 3.5 ms after the UDP datagram is sent, which
we’ll see in Chapter 7 is similar to the round-trip times for Ping replies.

84

6.6 4.4BSD Processing of ICMP Messages
Since ICMP covers such a wide range of conditions, from fatal errors to informational messages, each
ICMP message is handled differently, even within a given implementation. Figure 6.12 is a redo of
Figure 6.3, showing the handling performed by 4.4BSD for each of the possible ICMP messages.

If the final column specifies the kernel, that ICMP message is handled by the kernel. If the final
column specifies "user process", then that message is passed to all user processes that have registered
with the kernel to read received ICMP messages. If there are none of these user processes, the message
is silently discarded. (These user processes also receive a copy of all the other ICMP messages, even
those handled by the kernel, but only after the kernel has processed the message.) Some messages are
completely ignored. Finally, if the final column is a string in quotes, that is the Unix error message
corresponding to that condition. Some of these errors, such as TCP’s handling of a source quench,
we’ll cover in later chapters.

85

type code Description Handled by
0 0 echo reply user process
destination unreachable:
0 "No route to host"
network unreachable
host unreachable
2 "Connection refused"
protocol unreachable
3 "Connection refused"
port unreachable
4 "Message too long"
fragmentation needed but DF bit set
source route failed
destination network unknown
3 destination host unknown
source host isolated (obsolete)
dest. network administratively prohibited
dest. host administratively prohibited
network unreachable for TOS
host unreachable for TOS
13 (ignored)
communication administratively prohibited
14 (ignored)
host precedence violation
15 (ignored)
precedence cutoff in effect
4 0 source quench kernel for TCP, ignored by UDP
redirect:
0 kernel updates routing table
redirect for network
5 redirect for host
redirect for type-of-service and network
redirect for type-of-service and host
8 0 echo request kernel generates reply
9 0 router advertisement user process
10 0 router solicitation user process
time exceeded:
0 user process
11 TTL equals 0 during transit
1 user process
TTL equals 0 during reassembly
parameter problem:
0 "Protocol not available"
12 IP header bad (catchall error)
1 "Protocol not available"
required option missing
13 0 timestamp request kernel generates reply
14 0 timestamp reply user process
15 0 information request (obsolete) (ignored)
16 0 information reply (obsolete) user process
17 0 address mask request kernel generates reply
18 0 address mask reply user process

86

Figure 6.12 Handling of the ICMP message types by 4.4BSD.

6.7 Summary
This chapter has been a look at the Internet Control Message Protocol, a required part of every
implementation. Figure 6.3 lists all the ICMP message types, most of which we’ll discuss later in the
text.

We looked at the ICMP address mask request and reply and the timestamp request and reply in detail.
These are typical of the request-reply messages. Both have an identifier and sequence number in the
ICMP message. The sending application stores a unique value in the identifier field, to distinguish
between replies for itself and replies for other processes. The sequence number field lets the client
match replies with requests.

We also saw the ICMP port unreachable error, a common ICMP error. This let us examine the
information returned in an ICMP error: the IP header and the next 8 bytes of the IP datagram that
caused the error. This information is required by the receiver of the ICMP error, to know more about
the cause of the error. Both TCP and UDP store the source and destination port numbers in the first 8
bytes of their headers for this reason.

Finally, we presented our first time line of tcpdump output, a presentation format we’ll use in later
chapters.

Exercises
6.1 At the end of Section 6.2 we listed five special conditions under which an ICMP error message is
not sent. What would happen if these five conditions weren’t followed and we sent a broadcast UDP
datagram to an unlikely port on the local cable?

6.2 Read the Host Requirements RFC [Braden 1989a] to see if the generation of an ICMP port
unreachable is a "must," "should," or "may." What section and page is this found on?

6.3 Read RFC 1349 [Almquist 1992] to see how the IP type-of-service field (Figure 3.2) should be set
by ICMP.

6.4 If your system provides the netstat command, use it to see what types of ICMP messages are
received and sent.

87

Ping Program
7.1 Introduction
The name "ping" is taken from the sonar operation to locate objects. The Ping program was written by
Mike Muuss and it tests whether another host is reachable. The program sends an ICMP echo request
message to a host, expecting an ICMP echo reply to be returned. (Figure 6.3 lists all the ICMP
message types.)

Normally if you can’t Ping a host, you won’t be able to Telnet or FTP to that host. Conversely, if you
can’t Telnet to a host. Ping is often the starting point to determine what the problem is. Ping also
measures the round-trip time to the host, giving us some indication of how "far away" that host is.

In this chapter we’ll use Ping as a diagnostic tool and to further explore ICMP. Ping also gives us an
opportunity to examine the IP record route and timestamp options. Chapter II of [Stevens 1990]
provides the source code for the Ping program.

Years ago we could make the unqualified statement that if we can’t Ping a host, we can’t Telnet or FTP to that host. With
the increased awareness of security on the Internet, routers that provide access control lists, and firewall gateways,
unqualified statements like this are no longer true. Reachability ol a given host may depend not only on reachability at the
IP layer, but also on what protocol is being used, and the port numbers involved. Ping may show a host as being
unreachable, yet we might be able to Telnet to port 25 (the mail server).

7.2 Ping Program
We call the ping program that sends the echo requests the client, and the host being pinged the
server. Most TCP/IP implementations support the Ping server directly in the kernel-the server is not a
user process. (The two ICMP query services that we described in Chapter 6, the address mask and
timestamp requests, are also handled directly by the kernel.)

Figure 7.1 Format of ICMP message for echo request and echo reply.

As with other ICMP query messages, the server must echo the identifier and sequence number fields.
Also, any optional data sent by the client must be echoed. These are presumably of interest to the
client.

Unix implementations of ping set the identifier field in the ICMP message to the process ID of the
sending process. This allows ping to identify the returned responses if there are multiple instances of
ping running at the same time on the same host.

The sequence number starts at 0 and is incremented every time a new echo request is sent. ping prints
the sequence number of each returned packet, allowing us to see if packets are missing, reordered, or
duplicated. IP is a best effort datagram delivery service, so any of these three conditions can occur.

88

Historically the ping program has operated in a mode where it sends an echo request once a second,
printing each echo reply that is returned. Newer implementations, however, require the -s option to
operate this way. By default, these newer implementations send only a single echo request and output
"host is alive" if an echo reply is received, or "no answer" if no reply is received within 20 seconds.

LAN Output
ping output on a LAN normally looks like the following:

bsdi % ping svr4
PING svr4 (140.252.13.34): 56 data bytes
64 bytes from 140.252.13.34: icmp_seq=0 ttl=255 time=0 ms
64 bytes from 140.252.13.34: icmp_seq=l ttl=255 time=0 ms
^? type interrupt key to stop
--- svr4 ping statistics ---
8 packets transmitted, 8 packets received, 0% packet loss
round-trip min/avg/max = 0/0/0 ms

When the ICMP echo reply is returned, the sequence number is printed, followed by the TTL, and the
round-trip time is calculated. (TTL is the time-to-live field in the IP header. The current BSD ping
program prints the received TTL each time an echo reply is received-some implementations don’t do
this. We examine the usage of the TTL in Chapter 8 with the traceroute program.)

As we can see from the output above, the echo replies were returned in the order sent (0, 1,2, and so
on).

ping is able to calculate the round-trip time by storing the time at which it sends the echo request in the
data portion of the ICMP message. When the reply is returned it subtracts this value from the current
time. Notice that on the sending system, bsdi, the round-trip times are all calculated as 0 ms. This is
because of the low-resolution timer available to the program. The BSD/386 Version 0.9.4 system only
provides a IO-ms timer. (We talk more about this in Appendix B.) We’ll see later that when looking at
the tcpdump output from this ping example on a system with a finer resolution clock (the Sun) the
time difference between the ICMP echo request and its echo reply is just under 4 ms.

The first line of output contains the IP address of the destination host, even though we specified its
name (svr4). This implies that the name has been converted to the IP address by a resolver. We
examine resolvers and the DNS in Chapter 14. For now realize that if we type a ping command, and
a few seconds pass before the first line of output with the IP address is printed, this is the time required
for the DNS to determine the IP address corresponding to the hostname. Figure 7.2 shows the tcpdump
output for this example.

89

1 0.0 bsdi > svr4: icmp: echo request
2 0.003733 (0.0037) svr4 > bsdi: icmp: echo reply
3 0.998045 (0.9943) bsdi > svr4: icmp: echo request

Figure 7.2 ping output across a LAN.

The time between sending the echo request and receiving the echo reply is always 3.7 ms. We can also
see that echo requests are sent approximately 1 second apart.

Often the first round-trip time is larger than the rest. This occurs if the destination’s hardware address
isn’t in the ARP cache of the sender. As we saw in Chapter 4, sending an ARP request and getting the
ARP reply can take a few milliseconds before the first echo request can be sent. The following
example shows this:

sun % arp -a make sun ARP cache is empty

sun % ping svr4
PING svr4: 56 data bytes
64 bytes from svr4 (140.252.13,34): icmp_seq=0. time=7. ms
64 bytes from svr4 (140.252.13.34): icmp_seq=1. time=4. ms
---svr4 PING Statistics---
round-trip (ms) min/avg/max = 4/4/7

The additional 3 ms in the first RTT is probably for the ARP request and reply

This example was run on the host sun, which provides a timer with microsecond resolution, but the
ping program prints the round-trip times with only millisecond resolution. The earlier example, run
under BSD/386 Version 0.9.4, printed the round-trip times as 0 ms, since the available timer provided
only IO-ms accuracy. The following output is from BSD/386 Version 1.0, which provides a timer with

90

microsecond resolution and a version of ping that prints the higher resolution.

bsdi % ping svr4
PING svr4 (140.252.13.34): 56 data bytes
64 bytes from 140.252.13.34: icmp_seq=0 ttl=255 time=9.304 ms
--- svr4 ping statistics ---
round-trip min/avg/max = 6.079/6.880/9.304 ms

WAN Output
On a wide area network the results can be quite different. The following example was captured on a
weekday afternoon, a time when the Internet is normally busy:

gemini % ping vangogh.cs.berkeley.edu
PING vangogh.cs.berkeley.edu: 56 data bytes
64 bytes from (128.32.130.2): icmp_seq=0. time=660. ms
----vangogh.CS.Berkeley.EDU PING Statistics----

Either the echo requests or the echo replies for sequence numbers 1, 2, 3, 4, 6, 10, II, 12, and 13 were
lost somewhere. Note also the large variance in the round-trip times. (This high packet loss rate of
52% is an anomaly. This is not normal for the Internet, even on a weekday afternoon.)

It is also possible across WANs to see packets duplicated (the same sequence number printed two or
more times), and to see packets reordered (sequence number N +1 printed before sequence number N).

Hardwired SLIP Links
Let’s look at the round-trip times encountered over SLIP links, since they often run at slow
asynchronous speeds, such as 9600 bits/sec or less. Recall our serial line throughput calculations in
Section 2.10. For this example we’ll set the speed of the hardwired SLIP link between hosts bsdi and
slip to 1200 bits/sec.

We can estimate the round-trip time as follows. First, notice from the example Ping output shown
earlier that by default it sends 56 bytes of data in the ICMP message. With a 20-byte IP header and an
8-byte ICMP header this gives a total IP datagram size of 84 bytes. (We can verify this by running
tcpdump -e and seeing the Ethernet frame sizes.) Also, from Section 2.4 we know that at least two
additional bytes are added: the END byte at the beginning and end of the datagram. It’s also possible

91

for additional bytes to be added by the SLIP framing, but that depends on the value of each byte in the
datagram. At 1200 bits/sec with 8 bits per byte, I start bit, and I stop bit, the rate is 120 bytes per
second, or 8.33 ms per byte. Our estimate is then (86 x 8.33 x 2), or 1433 ms. (The multiplier of 2 is
because we are calculating the round-trip time.)

The following output verifies our calculation:

svr4 % ping -a slip
PING slip: 56 data bytes
64 bytes from slip (140.252.13.65): icmp_seq=0. time=1480. Ms
^?
----slip PING Statistics----

(The -s option is required for SVR4 to send one request every second.) The round-trip time is almost
1.5 seconds but the program is still sending out each ICMP echo request at 1-second intervals. This
means there are two outstanding echo requests (sent at time 0 and time 1) before the first reply comes
back (at time 1.480). That’s also why the summary line says one packet has been lost. It really hasn’t
been lost, it’s probably still on its way back.

We’ll return to this slow SLIP link in Chapter 8 when we examine the traceroute program.

Dialup SLIP Links
The conditions change with a dialup SLIP link since we now have modems on each end of the link.
The modems being used between the systems sun and netb provide what is called V.32 modulation
(9600 bits/sec), V.42 error control (also called LAP-M), and V.42bis data compression. This means
that our simple calculations, which were fairly accurate for a hardwired link where we knew all the
parameters, become less accurate.

Numerous factors are at work. The modems introduce some latency. The size of the packets may
decrease with the data compression, but the size may then increase to a multiple of the packet size
used by the error control protocol. Also the receiving modem can’t release received data bytes until the
cyclic redundancy character (the checksum) has been verified. Finally, we’re dealing with a
computer’s asynchronous serial interface on each end, and many operating systems read these
interfaces only at fixed intervals, or when a certain number of characters have been received.

As an example, we ping the host gemini from the host sun:

sun % ping gemini
PING gemini: 56 data bytes
64 bytes from gemini (140.252.1.11): icmp_seq=0. time=373. ms
64 bytes from gemini (140.252.1.11): icmp_seq=l. time=360. ms
64 bytes from gemini (140.252.1.11): icmp_seq=3. time-320. ms
64 bytes from gemini (140.252.1.11): icmp seq=5. time=310. ms
64 bytes from gemini (140.252.1.11): icmp seq=6. time==290. ms

92

64 bytes from gemini (140.252.1.11): icmp_seq-7. time=300. ms
64 bytes from gemini (140.252.1.11): icmp_seq=ll. time=280. ms
--gemini PING Statistics--

Note that the first RTT is not a multiple of 10 ms, but every other line is. If we run this numerous
times, we see this property every time. (This is not caused by the resolution of the clock on the host
sun, because we know that its clock provides millisecond resolution from the tests we run in Appendix
B.)

Also note that the first RTT is larger than the next, and they keep decreasing, and then they range
between 280 and 300 ms. If we let it run for a minute or two, the RTTs stay in this range, never going
below 260 ms. If we calculate the expected RTT at 9600 bits/sec (Exercise 7.2) we get 180 ms, so our
observed values are about 1.5 times the expected value.

If we run ping for 60 seconds and look at the average RTT it calculates, we find that with V.42 and
V.42bis our average is 277 ms. (This is better than the average printed for our preceding example,
because we ran it longer to amortize the longer RTTs at the beginning.) If we turn off just the V.42bis
data compression our average is 330 ms. If we turn off the V.42 error control (which also turns off the
V.42bis data compression) our average is 300 ms. These modem parameters do affect the RTTs, and
using the error control and data compression appears to be the best.

7.3 IP Record Route Option
The ping program gives us an opportunity to look at the IP record route (RR) option. Most versions
of ping provide the -R option that enables the record route feature. It causes ping to set the IP RR
option in the outgoing IP datagram (which contains the ICMP echo request message). This causes
every router that handles the datagram to add its IP address to a list in the options field. When the
datagram reaches the final destination, the list of IP addresses should be copied into the outgoing
ICMP echo reply, and all the routers on the return path also add their IP addresses to the list. When
ping receives the echo reply it prints the list of IP addresses.

As simple as this sounds, there are pitfalls. Generation of the RR option by the source host, processing
of the RR option by the intermediate routers, and reflection of the incoming RR list in an ICMP echo
request into the outgoing ICMP echo reply are all optional features. Fortunately, most systems today
do support these optional features, but some systems don’t reflect the IP list.

The biggest problem, however, is the limited room in the IP header for the list of IP addresses. We saw
in Figure 3.1 that the header length in the IP header is a 4-bit field, limiting the entire IP header to 15
32-bit words (60 bytes). Since the fixed size of the IP header is 20 bytes, and the RR option uses 3
bytes for overhead (which we describe below), this leaves 37 bytes (60-20-3) for the list, allowing up
to nine IP addresses. In the early days of the ARPANET, nine IP addresses seemed like a lot, but since
this is a round-trip list (in the case of the -R option for ping), it’s of limited use today. (In Chapter 8
we’ll look at the Traceroute tool for determining the route followed by a datagram.) Despite these
shortcomings, the record route option works and provides an opportunity to look in detail at the
handling of IP options. Figure 7.3 shows the general format of the RR option in the IP datagram.

93

Figure 7.3 General format of record route option in IP header.

Code is a I-byte field specifying the type of IP option. For the RR option its value is 7. Len is the total
number of bytes of the RR option, which in this case is 39. (Although it’s possible to specify an RR
option with less than the maximum size, ping always provides a 39-byte option field, to record up to
nine IP addresses. Given the limited room in the IP header for options, it doesn’t make sense to specify
a size less than the maximum.)

Ptr is called the pointer field. It is a 1-based index into the 39-byte option of where to store the next IP
address. Its minimum value is 4, which is the pointer to the first IP address. As each IP address is
recorded into the list, the value of ptr becomes 8, 12, 16, up to 36. After the ninth address is recorded
ptr becomes 40, indicating the list is full.

When a router (which by definition is multihomed) records its IP address in the list, which IP address
is recorded? It could be the address of the incoming interface or the outgoing interface. RFC 791
[Postel 1981a] specifies that the router records the outgoing IP address. We’ll see that when the
originating host (the host running ping) receives the ICMP echo reply with the RR option enabled, it
also records its incoming IP address in the list.

Normal Example
Let’s run an example of the RR option with the ping program. We’ll run ping on the host svr4 to
the host slip. One intermediate router (bsdi) will handle the datagram. The following output is
from svr4:

svr4 % ping -R slip
PING slip (140.252.13.65): 56 data bytes
RR: bsdi (140.252.13.66)
slip (140.252.13.65)
bsdi (140.252.13.35)
svr4 (140.252.13.34)
64 bytes from 140.252.13.65: icmp_seq=1 ttl=254 time=280 ms (same
route)
64 bytes from 140.252.13.65: icmp_seq=2 ttl=254 time=270 ms (same
route)
^?
--- slip ping statistics ---

Figure 7.4 shows the four hops that the packets take (two in each direction), and which hop adds which
IP address to the RR list.

94

Figure 7.4 ping with record route option.

The router bsdi adds a different IP address to the list in each direction. It always adds the IP address
of the outgoing interface. We can also see that when the ICMP echo reply reaches the originating
system (svr4) it adds the IP address of the incoming interface to the list.

We can also watch this exchange of packets from the host sun, running tcpdump with its -v option
(to see the IP options). Figure 7.5 shows the output.

svr4 > slip: icmp: echo request (ttl 32, id
35835,
1 0.0 optlen=40 RR{39}=RR(#0.0.0.0/0.0.0.0/0.0.0.0/
0.0.0.0/0.0.0.0/0.0.0.0/0.0.0.0/0.0.0.0/0.0.0.0}
EOL)
slip > svr4: icmp: echo reply (ttl 254, id 1976,
0.267746 optlen=40 RR{39}= RR(140.252.13.66/140.252.13.65/
2
(0.2677) 140.252.13.35/#0.0.0.0/0.0.0.0/0.0.0.0/0.0.0.0/
0.0.0.0/0.0.0.0} EOL)

Figure 7.5 tcpdump output of record route option.

The output optlen=40 indicates there are 40 bytes of option space in the IP header. (Recall that the
length of the IP header must be a multiple of 4 bytes.) RR{39} means the record route option is
present, and its length field is 39. The list of nine IP addresses is then shown, with a pound sign (#)
indicating which IP address is pointed to by the ptr field in the RR option header. Since we are
watching these packets on the host sun (see Figure 7.4) we only see the ICMP echo request with the
empty list, and the ICMP echo reply with three addresses in the list. We have deleted the remaining
lines in the tcpdump output, since they are nearly identical to what we show in Figure 7.5.

The notation EOL at the end of the record route information indicates the IP option "end of list" value
appeared. The EOL option has a value of 0. What’s happening is that 39 bytes of RR data are in the 40
bytes of option space in the IP header. Since the option space is set to 0 before the datagram is sent,
this final byte of 0 that follows the 39 bytes of RR data is interpreted as an EOL. That is what we want
to have happen. If there are multiple options in the option field of the IP header, and pad bytes are
needed before the next option starts, the other special character NOP ("no operation"), with a value of
1, can be used.

In Figure 7.5, SVR4 sets the TTL field of the echo request to 32, and BSD/386 sets it to 255. (It prints as 254 since the
router bsdi has already decremented it by one.) Newer systems are setting the TTL of ICMP messages to the maximum
(255).

It turns out that of the three TCP/IP implementations used by the author, both BSD/386 and SVR4 support the record route
option. That is, they correctly update the RR list when forwarding a datagram, and they correctly reflect the RR list from an
incoming ICMP echo request to the outgoing ICMP echo reply. SunOS 4.1.3, however, updates the RR list when
forwarding a datagram, but does not reflect the RR list. Solaris 2.x corrects this problem.

95

Abnormal Output
The following example was seen by the author and provides a starting point for our description of the
ICMP redirect message in Chapter 9. We ping the host aix on the 140.252.1 subnet (accessible
through the dialup SLIP connection on the host sun) with the record route option. We get the
following output, when run on the host slip:

slip % ping -R aix
PING aix (140.252.1.92): 56 data bytes
64 bytes from 140.252.1.92: icmp_seq-0 ttl=-251 time-650 ms
RR: bsdi (140.252.13.35)
sun (140.252.1.29)
netb (140.252.1.183)
aix (140.252.1.92)
gateway (140.252.1.4) why is this router used?
netb (140.252.1.183)
sun (140.252.13.33)
bsdi (140.252.13.66)
slip (140.252.13.65)
64 bytes from aix: icmp_seq=l ttl=251 time=610 ms (same route)
64 bytes from aix: icmp_seq=2 ttl=251 time"600 ms (same route)
^?
-- aix ping statistics -

We could have run this example from the host bsdi. We chose to run it from slip to see all nine IP
addresses in the RR list used.

The puzzle in this output is why the outgoing datagram (the ICMP echo request) went directly from
netb to aix, but the return (the ICMP echo reply) went from aix, through the router gateway,
before going to netb. What we’re seeing here is a feature of IP routing that we describe below.
Figure 7.6 shows the path of the datagrams.

Figure 7.6 ping with record route, showing IP routing feature.

The problem is that aix does not know to send IP datagrams destined for the subnet 140.252.13 to
netb. Instead, aix has a default entry in its routing table that tells it to send all datagrams to the
router gateway if it doesn’t have a particular route for the destination. The router gateway has
more routing knowledge than any of the hosts on the 140.252.1 subnet. (There are more than 150 hosts
on this Ethernet and instead of running a routing daemon on every one, each has a "default" entry that
points to the router gateway.)

96

An unanswered question here is why doesn’t gateway send an ICMP redirect (Section 9.5) to aix to
update its routing table? For some reason (perhaps that the datagram generating the redirect is an
ICMP echo request message) the redirect is not generated. But if we use Telnet and connect to the
daytime server on aix, the ICMP redirect is generated, and the routing table on aix is updated. If we
then execute ping with the record route option enabled, the route shows that the datagrams go from
netb to aix and back to netb, without the extra hop to the router gateway. We’ll look at these
ICMP redirects in more detail in Section 9.5.

7.4 IP Timestamp Option
The IP timestamp option is similar to the record route option. Figure 7.7 shows the format of the IP
timestamp option (compare with Figure 7.3).

Figure 7.7 General format of timestamp option in IP header.

The code field is 0x44 for the timestamp option. The two fields len and ptr are the same as for the
record route option: the total length of the option (normally 36 or 40) and a pointer to the next
available entry (5,9, 13, etc.).

The next two fields are 4-bit values: OF is the overflow field and FL is a flags field. The operation of
the timestamp option is driven by the flags field, as shown in Figure 7.8.

flags Description
0 Record only timestamps. This is what we show in Figure 7.7.
Each router records its IP address and its timestamp. There is room for only four of these pairs
1
in the options list.
The sender initializes the options list with up to four pairs of IP addresses and a 0 timestamp.
3
A router records its timestamp only if the next IP address in the list matches the router’s.

Figure 7.8 Meaning of the flags value for timestamp option.

If a router can’t add a timestamp because there’s no room left, it just increments the overflow field.

The preferred value for the timestamps is the number of milliseconds past midnight, UTC, similar to
the ICMP timestamp request and reply (Section 6.4). If this format is not available to a router, it can
insert whatever time representation that it uses, but must then turn on the high-order bit of the
timestamp to indicate the nonstandard value.

Given the limitations that we encountered with the record route option, things get worse with the
timestamp option. If we record both IP addresses and timestamps (a flags of 1), we can store only four
of these pairs. Recording only timestamps is next to useless because we have no indication regarding
which timestamp corresponds to which router (unless we have a fixed topology that never changes). A
flags of 3 is better, as we can then select which routers insert their timestamp. A more fundamental

97

problem is that you probably have no control over how accurate the timestamp is at any given router.
This makes it fruitless to try to measure hop times between routers using this IP option. We’ll see that
the traceroute program (Chapter 8) provides a better way of measuring hop times between routers.

7.5 Summary
The ping program is the basic connectivity test between two systems running TCP/IP. It uses the
ICMP echo request and echo reply messages and does not use a transport layer (TCP or UDP). The
Ping server is normally part of the kernel’s ICMP implementation.

We looked at the normal ping output for a LAN, WAN, and SLIP links (dialup and hardwired), and
performed some serial line throughput calculations for a dedicated SLIP link. ping also let us
examine and use the IP record route option. We used this IP option to see how default routes are often
used, and will return to this topic in Chapter 9. We also looked at the IP timestamp option, but it is of
limited practical use.

Exercises
7.1 Draw a time line for the ping output for the SLIP link in Section 7.2.

7.2 Calculate the RTT if the SLIP link between bsdi and slip is set to 9600 bits/sec. Assume the
default of 56 bytes of data.

7.3 The current BSD ping program allows us to specify a pattern for the data portion of the ICMP
message. (The first 8 bytes of the data portion are not filled with the pattern, since the time at which
the packet is sent is stored here.) If we specify a pattern of 0xc0, recalculate the answer to the
previous exercise. (Hint: Reread Section 2.4.)

7.4 Does the use of compressed SLIP (CSLIP, Section 2.5) affect the ping times that we observed in
Section 7.2?

7.5 Examine Figure 2.4. Do you expect any difference between a ping of the loopback address, versus
a ping of the host’s Ethernet address?

98

Traceroute Program
8.1 Introduction
The Traceroute program, written by Van Jacobson, is a handy debugging tool that allows us to further
explore the TCP/IP protocols. Although there are no guarantees that two consecutive IP datagrams
from the same source to the same destination follow the same route, most of the time they do.
Traceroute lets us see the route that IP datagrams follow from one host to another. Traceroute also lets
us use the IP source route option.

The manual page states: "Implemented by Van Jacobson from a suggestion by Steve Deering.
Debugged by a cast of thousands with particularly cogent suggestions or fixes from C. Philip Wood,
Tim Seaver, and Ken Adelman."

8.2 Traceroute Program Operation
In Section 7.3 we described the IP record route option (RR). Why wasn’t this used instead of
developing a new application? There are three reasons. First, historically not all routers have supported
the record route option, making it unusable on certain paths. (Traceroute doesn’t require any special or
optional features at any intermediate routers.)

Second, record route is normally a one-way option. The sender enables the option and the receiver has
to fetch all the values from the received IP header and somehow return them to the sender. In Section
7.3 we saw that most implementations of the Ping server (the ICMP echo reply function within the
kernel) reflect an incoming RR list, but this doubles the number of IP addresses recorded (the outgoing
path and the return path), which runs into the limit described in the next paragraph. (Traceroute
requires only a working UDP module at the destination-no special server application is required.)

The third and major reason is that the room allocated for options in the IP header isn’t large enough
today to handle most routes. There is room for only nine IP addresses in the IP header options field. In
the old days of the ARPANET this was adequate, but it is far too small nowadays.

Traceroute uses ICMP and the TTL field in the IP header. The TTL field (time-to-live) is an 8-bit field
that the sender initializes to some value. The recommended initial value is specified in the Assigned
Numbers RFC and is currently 64. Older systems would often initialize it to 15 or 32. We saw in some
of the Ping examples in Chapter 7 that ICMP echo replies are often sent with the TTL set to its
maximum value of 255.

Each router that handles the datagram is required to decrement the TTL by either one or the number of
seconds that the router holds onto the datagram. Since most routers hold a datagram for less than a
second, the TTL field has effectively become a hop counter, decremented by one by each router.

RFC 1009 [Braden and Postel 1987] required a router that held a datagram for more than 1 second to
decrement the TTL by the number of seconds. Few routers implemented this requirement. The new
Router Requirements RFC [Almquist 1993] makes this optional, allowing a router to treat the TTL as
just a hop count.

The purpose of the TTL field is to prevent datagrams from ending up in infinite loops, which can
occur during routing transients. For example, when a router crashes or when the connection between
two routers is lost, it can take the routing protocols some time (from seconds to a few minutes) to
detect the lost route and work around it. During this time period it is possible for the datagram to end

99

up in routing loops. The TTL field puts an upper limit on these looping datagrams.

When a router gets an IP datagram whose TTL is either 0 or 1 it must not forward the datagram. (A
destination host that receives a datagram like this can deliver it to the application, since the datagram
does not have to be routed. Normally, however, no system should receive a datagram with a TTL of 0.)
Instead the router throws away the datagram and sends back to the originating host an ICMP "time
exceeded" message. The key to Traceroute is that the IP datagram containing this ICMP message has
the router’s IP address as the source address.

We can now guess the operation of Traceroute. It sends an IP datagram with a TTL of 1 to the
destination host. The first router to handle the datagram decrements the TTL, discards the datagram,
and sends back the ICMP time exceeded. This identifies the first router in the path. Traceroute then
sends a datagram with a TTL of 2, and we find the IP address of the second router. This continues until
the datagram reaches the destination host. But even though the arriving IP datagram has a TTL of 1,
the destination host won’t throw it away and generate the ICMP time exceeded, since the datagram has
reached its final destination. How can we determine when we’ve reached the destination?

Traceroute sends UDP datagrams to the destination host, but it chooses the destination UDP port
number to be an unlikely value (larger than 30,000), making it improbable that an application at the
destination is using that port. This causes the destination host’s UDP module to generate an ICMP
"port unreachable" error (Section 6.5) when the datagram arrives. All Traceroute needs to do is
differentiate between the received ICMP messages-time exceeded versus port unreachable-to know
when it’s done.

The Traceroute program must be able to set the TTL field in the outgoing datagram. Not all
programming interfaces to TCP/IP support this, and not all implementations support the capability, but
most current systems do, and are able to run Traceroute. This programming interface normally requires
the user to have superuser privilege, meaning it may take special privilege to run it on your host.

8.3 LAN Output
We’re now ready to run traceroute and see the output. We’ll use our simple internet (see the figure on
the inside front cover) going from svr4 to slip, through the router bsdi. The hardwired SLIP link
between bsdi and slip is 9600 bits/sec.

svr4 % traceroute slip
traceroute to slip (140.252.13.65), 30 hops max. 40 byte packets
1 bsdi (140.252.13.35) 20 ms 10 ms 10 ms
2 slip (140.252.13.65) 120 ms 120 ms 120 ms

The first unnumbered line of output gives the name and IP address of the destination and indicates that
traceroute won’t increase the TTL beyond 30. The datagram size of 40 bytes allows for the 20-byte IP
header, the 8-byte UDP header, and 12 bytes of user data. (The 12 bytes of user data contain a
sequence number that is incremented each time a datagram is sent, a copy of the outgoing TTL, and
the time at which the datagram was sent.)

The next two lines in the output begin with the TTL, followed by the name of the host or router, and
its IP address. For each TTL value three datagrams are sent. For each returned ICMP message the
round-trip time is calculated and printed. If no response is received within 5 seconds for any of the
three datagrams, an asterisk is printed instead and the next datagram is sent. In this output the first
three datagrams had a TTL of 1 and the ICMP messages were returned in 20, 10, and 10 ms. The next

100

three datagrams were sent with a TTL of 2 and the ICMP messages were returned 120 ms later. Since
the TTL of 2 reached the final destination, the program then stopped.

The round-trip times are calculated by the traceroute program on the sending host. They are the
total RTTs from the traceroute program to that router. If we’re interested in the per-hop time we
have to subtract the value printed for TTL N from the value printed for TTL N+1.

Figure 8.1 shows the tcpdump output for this run. As we might have guessed, the reason that the first
probe packet to bsdi had an RTT of 20 ms and the next two had an RTT of 10 ms was because of an
ARP exchange, tcpdump shows this is indeed the case.

The destination UDP port starts at 33435 and is incremented by one each time a datagram is sent. This
starting port number can be changed with a command-line option. The UDP datagram contains 12
bytes of user data, which we calculated earlier when traceroute output that it was sending 40-byte
datagrams.

Next, tcpdump prints the comment [ttl 1] when the IP datagram has a TTL of 1. It prints a
message like this when the TTL is 0 or 1, to warn us that something looks funny in the datagram. Here
we expect to see the TTL of 1, but with some other application it could be a warning that the datagram
might not get to its final destination. We should never see a datagram passing by with a TTL of 0,
unless the router that put it on the wire is broken.

101

1 0.0 arp who-has bsdi tell svr4
0.000586
2 arp reply bsdi is-at 0:0:c0:6f:2d:40
(0.0006)
0.003067
3 svr4.42804 > slip.33435; udp 12 [ttl 1]
(0.0025)
0.004325
4 bsdi > svr4: icmp: time exceeded in-transit
(0.0013)
0.069810
5 svr4.42804 > slip.33436: udp 12 [ttl 1]
(0.0655)
0.071149
(0.0013)
0.085162
7 svr4.42804 > slip.33437: udp 12 [ttl 1]
(0.0140)
0.086375
(0.0012)
0.118608
9 svr4.42804 > slip.33438: udp 12
(0.0322)
0.226464 slip > svr4: icmp: slip udp port 33438
10
(0.1079) unreachable
0.287296
11 svr4.42804 > slip.33439: udp 12
(0.0608)
12
0.409504
13 svr4.42804 > slip.33440: udp 12
(0.0143)
14

Figure 8.1 tcpdump output for traceroute example from svr4 to slip.

The ICMP message "time exceeded in transit" is what we expect to see from the router bsdi, since it
will decrement the TTL to 0. The ICMP message comes from the router even though the IP datagram
that was thrown away was going to slip.

There are two different ICMP "time exceeded" messages (Figure 6.3), each with a different code field
in the ICMP message. Figure 8.2 shows the format of this ICMP error message.

102

Figure 8.2 ICMP time exceeded message.

The one we’ve been describing is generated when the TTL reaches 0, and is specified by a code of 0.

It’s also possible for a host to send an ICMP "time exceeded during reassembly" when it times out
during the reassembly of a fragmented datagram. (We talk about fragmentation and reassembly in
Section 11.5.) This error is specified by a code of 1.

Lines 9-14 in Figure 8.1 correspond to the three datagrams sent with a TTL of 2. These reach the final
destination and generate an ICMP port unreachable message.

It is worthwhile to calculate what the round-trip times should be for the SLIP link, similar to what we
did in Section 7.2 when we set the link to 1200 bits/sec for the Ping example. The outgoing UDP
datagram contains 12 bytes of data, 8 bytes of UDP header, 20 bytes of IP header, and 2 bytes (at
least) of SLIP framing (Section 2.4) for a total of 42 bytes. Unlike Ping, however, the size of the return
datagrams changes. Recall from Figure 6.9 that the returned ICMP message contains the IP header of
the datagram that caused the error and the first 8 bytes of data that followed that IP header (which is a
UDP header in the case of traceroute). This gives us a total of 20+8+20+8+2, or 58 bytes. With a data
rate of 960 bytes/sec the expected RTT is (42 + 58/960) or 104 ms. This corresponds to the IIO-ms
value measured on svr4.

The source port number in Figure 8.1 (42804) seems high. traceroute sets the source port number
of the UDP datagrams that it sends to the logical-OR of its Unix process ID with 32768. In case
traceroute is being run multiple times on the same host, each process looks at the source port
number in the UDP header that’s returned by ICMP, and only handles those messages that are replies
to probes that it sent.

There are several points to note with traceroute. First, there is no guarantee that the route today
will be in use tomorrow, or even that two consecutive IP datagrams follow the same route. If a route
changes while the program is running you’ll see it occur because traceroute prints the new IP
address for the given TTL if it changes.

Second, there is no guarantee that the path taken by the returned ICMP message retraces the path of
the UDP datagram sent by traceroute. This implies that the round-trip times printed may not be a
true indication of the outgoing and returning datagram times. (If it takes 1 second for the UDP
datagram to travel from the source to a router, but 3 seconds for the ICMP message to travel a different
path back to the source, the printed round-trip time is 4 seconds.)

Third, the source IP address in the returned ICMP message is the IP address of the interface on the
router on which the UDP datagram arrived. This differs from the IP record route option (Section 7.3),
where the IP address recorded was the outgoing interface’s address. Since every router by definition
has two or more interfaces, running traceroute from host A to host B can generate different output
than from host B to host A. Indeed, if we run traceroute from host slip to svr4 the output
becomes:

103

slip % traceroute svr4
traceroute to svr4 (140.252.13.34), 30 hops max, 40 byte packets
1 bsdi (140.252.13.66) 110 ms 110 ms 110 ms
2 svr4 (140.252.13.34) 110 ms 120 ms 110 ms

This time the IP address printed for host bsdi is 140.252.13.66, the SLIP interface, while previously it
was 140.252.13.35, the Ethernet interface. Since traceroute also tries to print the name associated
with an IP address, the names can change. (In our example both interfaces on bsdi have the same
name.)

Consider Figure 8.3. It shows two local area networks with a router connected to each LAN. The two
routers are connected with a point-to-point link. If we run traceroute from a host on the left LAN
to a host on the right LAN, the IP addresses found for the routers will be if1 and if3. But going the
other way will print the IP addresses if4 and if2. The two interfaces if2 and if3 share the same network
ID, while the other two interfaces have different network IDs.

Figure 8.3 Identification of interfaces printed by traceroute.

Finally, across wide area networks the traceroute output is much easier to comprehend if the IP
addresses are printed as readable domain names, instead of as IP addresses. But since the only piece of
information traceroute has when it receives the ICMP message is an IP address, it does a "reverse
name lookup" to find the name, given the IP address. This requires the administrator responsible for
that router or host to configure their reverse name lookup function correctly (which isn’t always the
case). We describe how an IP address is converted to a name using the DNS in Section 14.5.

8.4 WAN Output
The output shown earlier for our small internet is adequate for examining the protocols in action, but
more a realistic use of traceroute involves larger internets such as the worldwide Internet. Figure 8.4 is
from the host sun to the Network Information Center, the NIC.

sun % traceroute nic.ddn.mil
traceroute to nic.ddn.mil (192.112.36.5), 30 hops max, 40 byte
packets

1 netb.tuc.noao.edu (140.252.1.183) 218 ms 227 ms 233 ms
2 gateway.tuc.noao.edu (140.252.1.4) 233 ms 229 ms 204 ms

3 butch.telcom.arizona.edu (140.252.104.2) 204 ms 228 ms 234 ms
4 Gabby.Telcom.Arizona.EDU (128.196.128.1) 234 ms 228 ms 204 ms
5 NSIgate.Telcom.Arizona.EDU (192.80.43.3) 233 ms 228 ms 234 ms

6 JPLI.NSN.NASA.GOV (128.161.88.2) 234 ms 590 ms 262 ms
7 JPL3.NSN.NASA.GOV (192.100.15.3) 238 ms 223 ms 234 ms
8 GSFC3.NSN.NASA.GOV (128.161.3.33) 293 ms 318 ms 324 ms

104

9 GSFC8.NSN.NASA.GOV (192.100.13.8) 294 ms 318 ms 294 ms
10 SURA2.NSN.NASA.GOV (128.161.166.2) 323 ms 319 ms 294 ms
11 nsn-FIX-pe.sura.net (192.80.214.253) 294 ms 318 ms 294 ms
12 GSI.NSN.NASA.GOV (128.161.252.2) 293 ms 318 ms 324 ms

13 NIC.DDN.MIL (192.112.36.5) 324 ms 321 ms 324 ms

Figure 8.4 traceroute from host sun to nic.ddn.mil.

Since running this example for inclusion in the text, the NIC for non-DDN sites (i.e., nonmilitary) has
moved from nic.ddn.mil to rs.internic.net, the new "InterNIC."

Once the datagrams leave the tuc.noao.edu network they enter the telcom.arizona.edu
network. They then enter the NASA Science Internet, nsn.nasa.gov. The routers for TTLs 6 and 7 are
at the Jet Propulsion Laboratory (JPL). The network sura. net in the output for TTL II is the
Southeastern Universities Research Association Network. The name GSI at TTL 12 is Government
Systems, Inc., the operator of the NIC.

The second RTT for the TTL of 6 (590) is almost double the other two RTTs (234 and 262). This
illustrates the dynamics of IP routing. Something happened somewhere between the sending host and
this router that slowed down this datagram. Also, we can’t tell if it was the outbound datagram that got
held up or the return ICMP error.

The RTT for the first probe with a TTL of 3 (204) is less than the RTT for the first probe with a TTL
of 2 (233). Since each printed RTT is the total time from the sending host to that router, this can (and
does) happen. The example in Figure 8.5 is from the host sun to the author’s publisher.

sun % traceroute aw.com
traceroute to aw.com (192.207.117.2), 30 hops max, 40 byte packets

4 Gabby.Telcom.Arizona. EDU (128.196.128.1) 264 ms 228 ms 234 ms
5 Westgate.Telcom.Arizona. EDO (192.80.43.2) 234 ms 228 ms 234 ms
6 uu-ua.AZ.westnet.net (192.31.39.233) 263 ms 258 ms 264 ms
7 enssl42.UT.westnet.net (192.31.39.21) 263 ms 258 ms 264 ms
8 t3-2.Denver-cnss97.t3.ans.net (140.222.97.3) 293 ms 288 ms 275 ms
10 t3-1.St-Louis-cnss80.t3.ans.net (140.222.80.2) 282 ms 288 ms 294
ms
11 t3-1.Chicago-cnss24.t3.ans.net (140.222.24.2) 293 ms 288 ms 294
ms
12 t3-2.Cleveland-cnss40.t3.ans.net (140.222.40.3) 294 ms 288 ms 294
ms
13 t3-1.New-York-cnss32.t3.ans.net (140.222.32.2) 323 ms 318 ms 324
ms
14 t3-1.Washington-DC-cnss56.t3.ans.net (140.222.56.2) 323 ms 318 ms
324 ms
324 ms
16 t3-0.enssl36.t3.ans.net (140.222.136.1) 323 ms 318 ms 324 ms

105

17 Washington.DC.ALTER.NET (192.41.177.248) 323 ms 377 ms 324 ms
18 Boston.MA.ALTER.NET (137.39.12.2) 324 ms 347 ms 324 ms
19 AW-gw.ALTER.NET (137.39.62.2) 353 ms 378 ms 354 ms
20 aw.com (192.207.117.2) 354 ms 349 ms 354 ms

Figure 8.5 traceroute from host sun.tuc.noao.edu to aw.com.

This time the datagrams enter the regional network westnet.net (TTLs 6 and 7) after leaving the
telcom.arizona.edu network. They then enter the NSFNET backbone, t3.ans.net, which is
run by Advanced Network & Services. (T3 is the common abbreviation for the 45 Mbits/sec phone
lines used by the backbone.) The final network is alter.net, the connection point to the Internet
for aw.corn.

8.5 IP Source Routing Option
Normally IP routing is dynamic with each router making a decision about which next-hop router to
send the datagram to. Applications have no control of this, and are normally not concerned with it. It
takes tools such as Traceroute to figure out what the route really is.

The idea behind source routing is that the sender specifies the route. Two forms are provided:

Strict source routing. The sender specifies the exact path that the IP datagram must follow. If a
router encounters a next hop in the source route that isn’t on a directly connected network, an
ICMP "source route failed" error is returned.
Loose source routing. The sender specifies a list of IP address that the datagram must traverse, but
the datagram can also pass through other routers between any two addresses in the list.

Traceroute provides a way to look at source routing, as we can specify an option allowing us to force a
source route, and see what happens.

Some of the publicly available Traceroute source code packages contain patches to specify loose
source routing. But the standard versions normally don’t include this option. A comment in the patches
is that "Van Jacobson’s original traceroute (spring 1988) supported this feature, but he removed it due
to pressure from people with broken gateways." For the examples shown in this section, the author
installed these patches and modified them to allow both loose and strict source routing.

Figure 8.6 shows the format of the source route option.

Figure 8.6 General format of the source route option in the IP header.

This format is nearly identical to the format of the record route option that we showed in Figure 7.3.
But with source routing we have to fill in the list of IP addresses before sending the IP datagram, while
with the record route option we allocate room and zero out the list of IP addresses, letting the routers
fill in the next entry in the list. Also, with source routing we only allocate room for and initialize the
number of IP addresses required, normally fewer than nine. With the record route option we allocated
as much room as we could, for up to nine addresses.

106

The code is 0x83 for loose source routing, and 0x89 for strict source routing. The len and ptr fields
are identical to what we described in Section 7.3.

The source route options are actually called "source and record route" (LSRR and SSRR, for loose and
strict) since the list of IP addresses is updated as the datagram passes along the path. What happens is
as follows:

The sending host takes the source route list from the application, removes the first entry (it
becomes the destination address of the datagram), moves all the remaining entries left by one
entry (where left is as in Figure 8.6), and places the original destination address as the final entry
in the list. The pointer still points to the first entry in the list (e.g., the value of the pointer is 4).
Each router that handles the datagram checks whether it is the destination address of the
datagram. If not, the datagram is forwarded as normal. (In this case loose source routing must
have been specified, or we wouldn’t have received the datagram.)
If the router is the destination, and the pointer is not greater than the length, then (1) the next
address in the list (where ptr points) becomes the destination address of the datagram, (2) the IP
address corresponding to the outgoing interface replaces the source address just used, and (3) the
pointer is incremented by 4.

This is best explained with an example. In Figure 8.7 we assume that the sending application on host S
sends a datagram to D, specifying a source route of R1, R2, and R3.

Figure 8.7 Example of IP source routing.

In this figure the pound sign (#) denotes the pointer field, which assumes the values of 4, 8, 12, and 16.
The length field will always be 15 (three IP addresses plus 3 bytes of overhead). Notice how the
destination address of the IP datagram changes on every hop.

When an application receives data that was source routed, it should fetch the value of the received
route and supply a reversed route for sending replies.

The Host Requirements RFC specifies that a TCP client must be able to specify a source route, and
that a TCP server must be able to receive a source route, and use the reverse route for all segments on
that TCP connection. If the TCP server later receives a different source route, that newer source route
overrides the earlier one.

traceroute Examples with Loose Source Routing
The -g option to traceroute lets us specify intermediate routers to be used with loose source
routing. This option can be specified up to eight times. (The reason this is eight and not nine is that the
programming interface being used requires that the final entry be the destination.)

Recall from Figure 8.4 that the route to the NIC,nic.ddn.mil, was through the NASA Science
Internet. In Figure 8.8 we force the datagrams to pass through the NSFNET instead by specifying the
router enssl42.UT.westnet.net (192.31.39.21) as an intermediate router:

107

sun % traceroute -g 192.31.39.21 nic.ddn.mil
traceroute to nic.ddn.mil (192.112.36.5), 30 hops max, 40 byte
packets

4 enssl42.UT.westnet.net (192.31.39.21) 294 ms 288 ms 295 ms
7 t3-1.St-Louis-cnss80.t3.ans.net (140.222.80.2) 294 ms 318 ms 294
ms
8 * t3-1.Chicago-cnss24.t3.ans.net (140.222.24.2) 318 ms 295 ms
9 t3-2.Cleveland-cnss40.t3.ans.net (140.222.40.3) 319 ms 318 ms 324
ms
10 t3-1.New-York-cnss32.t3.ans.net (140.222.32.2) 324 ms 318 ms 324
ms
325 ms
325 ms
13 13-0. enssl45.t3.ans. net (140.222.145.1) 353 ms 348 ms 325 ms
14 nsn-FIX-pe.sura.net (192.80.214.253) 353 ms 348 ms 325 ms
15 GSI.NSN.NASA.GOV (128.161.252.2) 353 ms 348 ms 354 ms
16 NIC.DDN.MIL (192.112.36.5) 354 ms 347 ms 354 ms

Figure 8.8 traceroute to nic.ddn.mil with a loose source route through the NSFNET.

This time there appear to be 16 hops with an average RTT of around 350 ms, while the normal route
shown in Figure 8.4 had only 13 hops and an average RTT of around 322 ms. The default route
appears better. (There are also other decisions made when routes are established. Some are made on
the basis of the organizational and political boundaries of the networks involved.)

But we said there appear to be 16 hops, because a comparison of this output with our previous
example through the NSFNET (Figure 8.5) shows three missing routers in this example using loose
source routing. (These are probably caused by bugs in the router’s generation of ICMP time exceeded
errors in response to source routed datagrams.) The router gateway.tuc.noao.edu is missing
between netb and butch, and the routers Westgate.Telcom.Arizona.edu and
uu-ua.AZ.westnet.net are also missing between Gabby and enssl42.UT.westnet.net. There is
probably a software problem in these missing routers related to incoming datagrams with the loose
source routing option. There are really 19 hops between the source and the NIC, when using the
NSFNET. Exercise 8.5 continues the discussion of these missing routers.

This example also illustrates another problem. On the command line we have to specify the
dotted-decimal IP address of the router enssl42.UT.westnet.net instead of its name. This is
because the reverse name lookup (return the name, given the IP address. Section 14.5), associates the
name with the IP address, but the forward lookup (given the name, return the IP address) fails. The
forward mapping and reverse mapping are two separate files in the DNS (Domain Name System) and
not all administrators keep the two synchronized with each other. It’s not uncommon to have one
direction work and the other direction fail.

108

Something that we haven’t seen before is the asterisk (*) printed for the first RTT for the TTL of 8.
This indicates that a timeout occurred and no response was received within 5 seconds for this probe.

Another point that we can infer from a comparison of this figure and Figure 8.4 is that the router
nsn-FIX-pe.sura.net is connected to both the NSFNET and the NASA Science Internet.

traceroute Examples with Strict Source Routing
The -G option in the author’s version of traceroute is identical to the -g option described earlier,
but the source route is strict instead of loose. We can use this to see what happens when an invalid
strict source route is specified. Recall from Figure 8.5 that the normal sequence of routers for
datagrams from the author’s subnet to the NSFNET is through netb, gateway, butch, and gabby.
(We’ve omitted the domain suffixes, .tuc.noao.edu and .telcom.arizona.edu, in all the output below to
make it easier to read.) We specify a strict source route that omits butch, trying to go directly from
gateway to gabby. We expect this to fail, as shown in Figure 8.9.

sun % traceroute -G netb -G gateway -G gabby westgate
traceroute to westgate (192.80.43.2), 30 hops max. 40 byte packets

1 netb (140.252.1.183) 272 ms 257 ms 261 ms
2 gateway (140.252.1.4) 263 ms 259 ms 234 ms
3 gateway (140.252.1.4) 263 ms !S * 235 ms !S

Figure 8.9 traceroute with a strict source route that fails.

The key here is the notation ! s following the RTTs for the TTL of 3. This indicates that
traceroute received an ICMP "source route failed" error message: a type of 3 and a code of 5 from
Figure 6.3. The asterisk for the second RTT for the TTL of 3 indicates no response was received for
that probe. This is what we expect, since it’s impossible for gateway to send the datagram directly to
gabby, because they’re not directly connected.

The reason that both TTLs 2 and 3 are from gateway is that the values for the TTL of 2 are from
gateway when it receives the datagram with an incoming TTL of 1. It discovers that the TTL has
expired before it looks at the (invalid) strict source route, and sends back the ICMP time exceeded.
The line with a TTL of 3 is received by gateway with an incoming TTL of 2, so it looks at the strict
source route, discovers that it’s invalid, and sends back the ICMP source route failed error.

Figure 8.10 shows the tcpdump output corresponding to this example. This output was collected on
the SLIP link between sun and netb. We had to specify the -v option for tcpdump to display the
source route information. This produces other output that we don’t need, such as the datagram ID,
which we’ve deleted. Also, the notation SSRR stands for "strict source and record route."

sun.33593 > netb.33435: udp 12 [ttl 1]
1 0.0 (optlen=16 SSRR{#gateway gabby westgate}
EOL)
0.270278
2 netb > sun: icmp: time exceeded in-transit
(0.2703)
sun.33593 > netb.33436: udp 12 [ttl 1]
0.284784
3 (optlen=16 SSRR{#gateway gabby westgate}
(0.0145)
EOL)

109

0.540338
(0.2556)
sun.33593 > netb.33437: udp 12 [ttl 1]
0.550062
5 (optlen=16 SSRR(#gateway gabby westgate}
(0.0097)
EOL)
0.810310
(0.2602)
sun.33593 > netb.33438: udp 12 (ttl 2,
0.818030
7 optlen=16 SSRR(#gateway gabby westgate}
(0.0077)
EOL)
1.080337 gateway > sun: icmp: time exceeded
8
(0.2623) in-transit
sun.33593 > netb.33439: udp 12 (ttl 2,
1.092564
9 optlen=16 SSRR{#gateway gabby westgate}
(0.0122)
EOL)
10
(0.2578) in-transit
sun.33593 > netb.33440: udp 12 (ttl 2,
1.357382
11 optlen=16 SSRR(#gateway gabby westgate}
(0.0071)
EOL)
12
(0.2332) in-transit
sun.33593 > netb.33441: udp 12 (ttl 3,
1.598926
(0.0083)
EOL)
gateway > sun:
1.860341
14 icmp: gateway unreachable - source route
(0.2614)
failed
sun.33593 > netb.33442: udp 12 (ttl 3,
1.875230
(0.0149)
EOL)
sun.33593 > netb.33443: udp 12 (ttl 3,
6.876579
(5.0013)
EOL)
gateway > sun:
7.110518
17 icmp: gateway unreachable - source route
(0.2339)
failed

Figure 8.10 tcpdump output of traceroute with failed strict source route.

110

First note that each UDP datagram sent by sun has a destination of netb, not the destination host
(westgate). We described this with the example shown in Figure 8.7. Similarly, the other two
routers specified with the -G option (gateway and gabby) and the final destination (westgate)
become the SSRR option list on the first hop.

We can also see from this output that the timeout used by traceroute (the time difference between
lines 15 and 16) is 5 seconds.

traceroute Round Trips with Loose Source Routing
Earlier we said that there is no guarantee that the route from A to Bis the same as the route from B to
A. Other than having a login on both systems and running traceroute on each end, it’s hard to find
out if there is a difference in the two paths. Using loose source routing, however, we can determine the
route in both directions.

The trick is to specify loose source routing with the destination as the loose route, and the sending host
as the final destination. For example, on the host sun we can find the paths to and from the host
bruno.cs.colorado.edu (Figure 8.11) .

sun % traceroute -g bruno.cs.colorado.edu sun
traceroute to sun (140.252.13.33), 30 hops max, 40 byte packets

5 NSIgate.Telcom.Arizona.EDU (192.80.43.3) 294 ms 258 ms 234 ms
6 JPLI.NSN.NASA.GOV (128.161.88.2) 264 ms 258 ms 264 ms
7 JPL2.NSN.NASA.GOV (192.100.15.2) 264 ms 258 ms 264 ms
8 NCAR.NSN.NASA.GOV (128.161.97.2) 324 ms * 295 ms
9 cu-gw.ucar.edu (192.43.244.4) 294 ms 318 ms 294 ms
10 engr-gw.Colorado.EDU (128.138.1.3) 294 ms 288 ms 294 ms
11 bruno.cs.Colorado.edu (128.138.243.151) 293 ms 317 ms 294 ms
12 engr-gw-ot.cs.Colorado.edu (128.138.204.1) 323 ms 317 ms 384 ms
13 cu-gw.Colorado.EDU (128.138.1.1) 294 ms 318 ms 294 ms
14 enss.ucar.edu (192.43.244.10) 323 ms 318 ms 294 ms
16 t3-0.enssl42.t3.ans.net (140.222.142.1) 293 ms 288 ms 294 ms
18 Butch.Telcom.Arizona.EDU (128.196.128.88) 293 ms 317 ms 294 ms
21 sun.tuc.noao.edu (140.252.13.33) 534 ms 529 ms 564 ms

Figure 8.11 traceroute example showing unsymmetrical routing path.

The outbound path (TTLs 1-11) differs from the return path (TTLs 11-21), a good illustration that
Internet routing need not be symmetrical.

111

This output also illustrates the point we discussed with Figure 8.3. Compare the output for TTLs 2 and
19: both are for the router gateway.tuc.noao.edu, but the two IP addresses are different. Since
traceroute identifies the incoming interface, and since we’re passing through the router in two
different directions, once on the outbound path (TTL 2) and then on the return path (TTL 19), we
expect this. We see the same effect comparing TTLs 3 and 18, and TTLs 4 and 17.

8.6 Summary
Traceroute is an indispensable tool when working with a TCP/IP network. Its operation is simple: send
UDP datagrams starting with a TTL of 1, increasing the TTL by 1, to locate each router in the path. An
ICMP time exceeded is returned by each router when it discards the UDP datagram, and an ICMP port
unreachable is generated by the final destination.

We ran examples of traceroute on both LANs and WANs, and used it to examine IP source
routing. We used loose source routing to see if the route to a destination is the same as the return route
from that destination.

Exercises
8.1 What can happen if an IP implementation decrements the incoming TTL and then tests for equal to
0?

8.2 How does traceroute calculate the RTT? Compare this to the RTT calculation done by ping.

8.3 (This exercise and the next one are based on actual problems determined when traceroute was
being developed, and are from comments in the traceroute source code.) Assume there are three
routers (Rl, R2, and R3) between the source and destination and that the middle router (R2)
decrements the TTL but incorrectly forwards the IP datagram when the incoming TTL was 1. Describe
what happens. How can you see this occur when running traceroute?

8.4 Again assume there are three routers between the source and destination. This time the destination
host has a bug whereby it always uses the incoming TTL as the outgoing TTL of an ICMP message.
Describe what happens and how you would see this.

8.5 We can run tcpdump on the SLIP link between sun and netb when running the example from
Figure 8.8. If we specify the -v option we can see the TTL value of the returned ICMP messages.
Doing this shows the incoming TTL from netb to be 255, from butch it’s 253, from Gabby it’s 252,
and from enssl42.UT.westnet.net it’s 249. Does this give any additional information about
whether there really are some missing routers?

8.6 Both SunOS and SVR4 provide a version of ping with a -l option that provides a loose source
route. The manual pages state that it’s intended to be used with the -R option (which specifies the
record route option). If you have access to either of these systems, try these two options together.
What’s happening? If you can watch the datagrams with tcpdump, describe what’s going on.

8.7 Compare the ways ping and traceroute handle multiple instances of the client on the same
host.

8.8 Compare the ways ping and traceroute measure the round-trip time.

112

8.9 We said traceroute picks the starting UDP destination port number at 33435 and increments
this by one for each packet sent. In Section 1.9 we said ephemeral port numbers are normally between
1024 and 5000, making it unlikely that Traceroute’s destination port is in use on the destination host.
Is this still true under Solaris 2.2? (Hint: Read Section E.4.)

8.10 Read RFC 1393 [Malkin 1993b] for a proposed alternative way of determining the path to a
destination. What are its advantages and disadvantages?

113

IP Routing
9.1 Introduction
Routing is one of the most important functions of IP. Figure 9.1 shows a simplified view of the
processing done at the IP layer. Datagrams to be routed can be generated either on the local host or on
some other host. In the latter case this host must be configured to act as a router, or datagrams received
through the network interfaces that are not ours are dropped (i.e., silently discarded).

In Figure 9.1 we also show a routing daemon, which is normally a user process. The most common
daemons used on Unix systems are the programs routed and gated. (The term daemon means the
process is running "in the background," carrying out operations on behalf of the whole system.
Daemons are normally started when the system is bootstrapped and run as long as the system is up.)
The topics of which routing protocol to use on a given host, how to exchange routing information with
adjacent routers, and how the routing protocols work are complex and can fill an entire book of their
own. (Interested readers are referred to [Periman 1992] for many of the details.) We’ll look briefly at
dynamic routing and the Routing Information Protocol (RIP) in Chapter 10. Our main interest in the
current chapter is how a single IP layer makes its routing decisions.

The routing table that we show in Figure 9.1 is accessed frequently by IP (on a busy host this could
mean hundreds of times a second) but is updated much less frequently by a routing daemon (possibly
about once every 30 seconds). The routing table can also be updated when ICMP "redirect" messages
are received, something we’ll look at in Section 9.5, and by the route command. This command is
often executed when the system is bootstrapped, to install some initial routes. We’ll also use the
netstat command in this chapter to display the routing table.

Figure 9.1 Processing done at the IP layer.

9.2 Routing Principles
The place to start our discussion of IP routing is to understand what is maintained by the kernel in its
routing table. The information contained in the routing table drives all the routing decisions made by
IP.

In Section 3.3 we listed the steps that IP performs when it searches its routing table.

1. Search for a matching host address.
2. Search for a matching network address.
3. Search for a default entry. (The default entry is normally specified in the routing table as a
network entry, with a network ID of 0.)

A matching host address is always used before a matching network address.

114

The routing done by IP, when it searches the routing table and decides which interface to send a packet
out, is a routing mechanism. This differs from a routing policy, which is a set of rules that decides
which routes go into the routing table. IP performs the routing mechanism while a routing daemon
normally provides the routing policy.

Simple Routing Table
Let’s start by looking at some typical host routing tables. On the host svr4 we execute the netstat
command with the -r option to list the routing table and the -n option, which prints IP addresses in
numeric format, rather than as names. (We do this because some of the entries in the routing table are
for networks, not hosts. Without the -n option, the netstat command searches the file
/etc/networks for the network names. This confuses the discussion by adding another set of
names-network names in addition to hostnames.)

svr4 % netstat
-rn
Routing tables
Destination Gateway Flags Refcnt Use Interface
140.252.13.65 140.252.13.35 UGH 0 0 emd0
127.0.0.1 127.0.0.1 UH 1 0 lo0
default 140.252.13.33 UG 0 0 emd0
140.252.13.32 140.252.13.34 U 4 25043 emd0

The first line says for destination 140.252.13.65 (host slip) the gateway (router) to send the packet to
is 140.252.13.35 (bsdi). This is what we expect, since the host slip is connected to bsdi with a
SLIP link, and bsdi is on the same Ethernet as this host. There are five different flags that can be
printed for a given route.

U The route is up.
G The route is to a gateway (router). If this flag is not set, the destination is directly connected.
H The route is to a host, that is, the destination is a complete host address. If this flag is not set, the
route is to a network, and the destination is a network address: a net ID, or a combination of a net ID
and a subnet ID.
D The route was created by a redirect (Section 9.5).
M The route was modified by a redirect (Section 9.5).

The G flag is important because it differentiates between an indirect route and a direct route. (The G
flag is not set for a direct route.) The difference is that a packet going out a direct route has both the IP
address and the link-layer address specifying the destination (Figure 3.3). When a packet is sent out an
indirect route, the IP address specifies the final destination but the link-layer address specifies the
gateway (that is, the next-hop router). We saw an example of this in Figure 3.4. In this routing table
example we have an indirect route (the G flag is set) so the IP address of a packet using this route is the
final destination (140.252.13.65), but the link-layer address must correspond to the router
140.252.13.35.

It’s important to understand the difference between the G and H flags. The G flag differentiates
between a direct and an indirect route, as described above. The H flag, however, specifies that the
destination address (the first column of netstat output) is a complete host address. The absence of
the H flag means the destination address is a network address (the host ID portion will be 0). When the

115

routing table is searched for a route to a destination IP address, a host address entry must match the
destination address completely, while a network address only needs to match the network ID and any
subnet ID of the destination address. Also, some versions of the netstat command print all the host
entries first, followed by the network entries.

The reference count column gives the number of active uses for each route. A connection-oriented
protocol such as TCP holds on to a route while the connection is established. If we established a Telnet
connection between the two hosts svr4 and slip, we would see the reference count go to 1. With
another Telnet connection the reference count would go to 2, and so on.

The next column ("use") displays the number of packets sent through that route. If we are the only
users of the route and we run the ping program to send 5 packets, the count goes up by 5. The final
column, the interface, is the name of the local interface.

The second line of output is for the loopback interface (Section 2.7), always named lo0. The G flag is
not set, since the route is not to a gateway. The H flag indicates that the destination address (127.0.0.1)
is a host address, and not a network address. When the G field is not set, indicating a direct route, the
gateway column gives the IP address of the outgoing interface.

The third line of output is for the default route. Every host can have one or more default routes. This
entry says to send packets to the router 140.252.13.33 (sun) if a more specific route can’t be found.
This means the current host (svr4) can access other systems across the Internet through the router
sun (and its SLIP link), using this single routing table entry. Being able to establish a default route is
a powerful concept. The flags for this route (UG) indicate that it’s a route to a gateway, as we expect.

Here we purposely call sun a router and not a host because when it’s used as a default router, its IP
forwarding function is being used, not its host functionality.

The Host Requirements RFC specifically states that the IP layer must support multiple default routes.
Many implementations, however, don’t support this. When multiple default routes exist, a common
technique is to round robin among them. This is what Solaris 2.2 does, for example.

The final line of output is for the attached Ethernet. The H flag is not set, indicating that the destination
address (140.252.13.32) is a network address with the host portion set to 0. Indeed, the low-order 5
bits are 0 (Figure 3.11). Since this is a direct route (the G flag is not set) the gateway column specifies
the IP address of the outgoing interface.

Implied in this final entry, but not shown by the netstat output, is the mask associated with this
destination address (140.252.13.32). If this destination is being compared against the IP address
140.252.13.33, the address is first logically ANDed with the mask associated with the destination (the
subnet mask of the interface, 0xffffffe0, from Section 3.7) before the comparison. For a network
route to a directly connected network, the routing table mask defaults to the subnet mask of the
interface. But in general the routing table mask can assume any 32-bit value. A value other than the
default can be specified as an option to the route command.

The complexity of a host’s routing table depends on the topology of the networks to which the host
has access.

1. The simplest (but least interesting) case is a host that is not connected to any networks at all. The
TCP/IP protocols can still be used on the host, but only to communicate with itself! The routing
table in this case consists of a single entry for the loopback interface.
2. Next is a host connected to a single LAN, only able to access hosts on that LAN. The routing

116

table consists of two entries: one for the loopback interface and one for the LAN (such as an
Ethernet).
3. The next step occurs when other networks (such as the Internet) are reachable through a single
router. This is normally handled with a default entry pointing to that router.
4. The final step is when other host-specific or network-specific routes are added. In our example
the route to the host slip, through the router bsdi, is an example of this.

Let’s follow through the steps IP performs when using this routing table to route some example
packets on the host svr4.

1. Assume the destination address is the host sun, 140.252.13.33. A search is first made for a
matching host entry. The two host entries in the table (slip and localhost) don’t match, so a
search is made through the routing table again for a matching network address. A match is found
with the entry 140.252.13.32 (the network IDs and subnet IDs match), so the emd0 interface is
used. This is a direct route, so the link-layer address will be the destination address.
2. Assume the destination address is the host slip, 140.252.13.65. The first search through the
table, for a matching host address, finds a match. This is an indirect route so the destination IP
address remains 140.252.13.65, but the link-layer address must be the link-layer address of the
gateway 140.252.13.35, and the interface is emd0.
3. This time we’re sending a datagram across the Internet to the host aw.com (192.207.117.2). The
first search of the routing table for a matching host address fails, as does the second search for a
matching network address. The final step is a search for a default entry, and this succeeds. The
route is an indirect route through the gateway 140.252.13.33 using the interface emd0.
4. In our final example we send a datagram to our own host. There are four ways to do this, using
either the hostname, the host IP address, the loopback name, or the loopback IP address:

ftp svr4 ftp
140.252.13.34

ftp localhost
ftp 127.0.0.1

In the first two cases, the second search of the routing table yields a network match with
140.252.13.32, and the packet is sent down to the Ethernet driver. As we showed in Figure 2.4 it
will be seen that this packet is destined for the host’s own IP address, and the packet is sent to the
loopback driver, which sends it to the IP input queue.

In the latter two cases, specifying the name of the loopback interface or its IP address, the first
search of the routing table finds the matching host address entry, and the packet is sent to the
loopback driver, which sends it to the IP input queue.

In all four cases the packet is sent to the loopback driver, but two different routing decisions are
made.

Initializing a Routing Table
We never said how these routing table entries are created. Whenever an interface is initialized
(normally when the interface’s address is set by the ifconfig command) a direct route is
automatically created for that interface. For point-to-point links and the loopback interface, the route is
to a host (i.e., the H flag is set). For broadcast interfaces such as an Ethernet, the route is to that
network.

117

Routes to hosts or networks that are not directly connected must be entered into the routing table
somehow. One common way is to execute the route command explicitly from the initialization files
when the system is bootstrapped. On the host svr4 the following two commands were executed to
add the entries that we showed earlier:

route add default sun 1
route add slip bsdi 1

The third arguments (default and slip) are the destinations, the fourth argument is the gateway
(router), and the final argument is a routing metric. All that the route command does with this metric is
install the route with the G flag set if the metric is greater than 0, or without the G flag if the metric is
0.

Unfortunately, few systems agree on which start-up file contains the route commands. Under 4.4BSD and BSD/386 it is
/etc/netstart, under SVR4 it is /etc/inet/rc.inet, under Solaris 2.x it is /etc/rc2.d/S69inet, SunOS
4.1.x uses /etc/rc.local, and AIX 3.2.2 uses /etc/rc.net.

Some systems allow a default router to be specified in a file such as /etc/defaultrouter, and
this default is added to the routing table on every reboot.

Other ways to initialize a routing table are to run a routing daemon (Chapter 10) or to use the newer
router discovery protocol (Section 9.6).

A More Complex Routing Table
The host sun is the default router for all the hosts on our subnet, since it has the dialup SLIP link that
connects to the Internet (see the figure on the inside front cover).

sun % netstat -rn
Routing tables
140.252.13.65 140.252.13.35 UGH 0 171 le0
127.0.0.1 127.0.0.1 UH 1 766 lo0
140.252.1.183 140.252.1.29 UH 0 0 sl0
default 140.252.1.183 UG 1 2955 sl0
140.252.13.32 140.252.13.33 U 8 99551 le0

The first two entries are identical to the first two for the host svr4: a host-specific route to slip
through the router bsdi, and the loopback route.

The third line is new. It is a direct route (the G flag is not set) to a host (the H flag is set) and
corresponds to our point-to-point link, the SLIP interface. If we compare it to the output from the
ifconfig command,

118

sun % ifconfig sl0
sl0: flags=1051<UP,POINTOPOINT,RUNNING>
inet 140.252.1.29 -> 140.252.1.183 netmask ffffff00

we see that the destination address in the routing table is the other end of the point-to-point link (the
router netb) and the gateway address is really the local IP address of the outgoing interface
(140.252.1.29). (We said earlier that the gateway address printed by netstat for a direct route is the
local IP address of the interface to use.)

The default entry is an indirect route (G flag) to a network (no H flag), as we expect. The gateway
address is the address of the router (140.252.1.183, the other end of the SLIP link) and not the local IP
address of the SLIP link (140.252.1.29). Again, this is because it is an indirect route, not a direct route.

We should also note that the third and fourth lines output by netstat (the ones with an interface of
sl0) are created by the SLIP software being used when the SLIP line is brought up, and deleted when
the SLIP link is brought down.

No Route to Destination
All our examples so far have assumed that the search of the routing table finds a match, even if the
match is the default route. What if there is no default route, and a match isn’t found for a given
destination?

The answer depends on whether the IP datagram being routed was generated on the host or is being
forwarded (e.g., we’re acting as a router). If the datagram was generated on this host, an error is
returned to the application that sent the datagram, either "host unreachable" or "network unreachable."
If the datagram was being forwarded, an ICMP host unreachable error is sent back to original sender.
We examine this error in the following section.

9.3 ICMP Host and Network Unreachable Errors
The ICMP "host unreachable" error message is sent by a router when it receives an IP datagram that it
cannot deliver or forward. (Figure 6.10 shows the format of the ICMP unreachable messages.) We can
see this easily on our network by taking down the dialup SLIP link on the router sun, and trying to
send a packet through the SLIP link from any of the other hosts that specify sun as the default router.

Older implementations of the BSD TCP/IP software generated either a host unreachable, or a network unreachable,
depending on whether the destination was on a local subnet or not. 4.4BSD generates only the host unreachable.

Recall from the netstat output for the router sun shown in the previous section that the routing
table entries that use the SLIP link are added when the SLIP link is brought up, and deleted when the
SLIP link is brought down. This means that when the SLIP link is down, there is no default route on
sun. But we don’t try to change all the other host’s routing tables on our small network, having them
also remove their default route. Instead we count on the ICMP host unreachable generated by sun for
any packets that it gets that it cannot forward.

We can see this by running ping on svr4, for a host on the other side of the dialup SLIP link (which
is down):

119

svr4 % ping gemini
ICMP Host Unreachable from gateway sun (140.252.13.33)
ICMP Host Unreachable from gateway sun (140.252.13.33)

Figure 9.2 shows the tcpdump output for this example, run on the host bsdi.

1 0.0 svr4 > gemini: icmp: echo request
2 0.00 (0.00) sun > svr4: icmp: host gemini unreachable
3 0.99 (0.99) svr4 > gemini: icmp: echo request
4 0.99 (0.00) sun > svr4: icmp: host gemini unreachable

Figure 9.2 ICMP host unreachable in response to ping.

When the router sun finds no route to the host gemini, it responds to the echo request with a host
unreachable.

If we bring the SLIP link to the Internet up, and try to ping an IP address that is not connected to the
Internet, we expect an error. What is interesting is to see how far the packet gets into the Internet,
before the error is returned:

sun % ping 192.82.148.1 this IP address is not connected to the
Internet
PING 192.82.148.1: 56 data bytes
ICMP Host Unreachable from gateway enss142.UT.westnet.net
(192.31.39.21) for icmp from sun (140.252.1.29) to 192.82.148.1

Looking at Figure 8.5 we see that the packet made it through six routers before detecting that the IP
address was invalid. Only when it got to the border of the NSFNET backbone was the error detected.
This implies that the six routers that forwarded the packet were doing so because of default entries,
and only when it reached the NSFNET backbone did a router have complete knowledge of every
network connected to the Internet. This illustrates that many routers can operate with just partial
knowledge of the big picture.

[Ford, Rekhter, and Braun 1993] define a top-level routing domain as one that maintains routing
information to most Internet sites and does not use default routes. They note that five of these top-level
routing domains exist on the Internet: the NFSNET backbone, the Commercial Internet Exchange
(CIX), the NASA Science Internet (NSI), SprintLink, and the European IP Backbone (EBONE).

9.4 To Forward or Not to Forward
We’ve mentioned a few times that hosts are not supposed to forward IP datagrams unless they have
been specifically configured as a router. How is this configuration done?

Most Berkeley-derived implementations have a kernel variable named ipforwarding, or some
similar name. (See Appendix E.) Some systems (BSD/386 and SVR4, for example) only forward
datagrams if this variable is nonzero. SunOS 4.1.x allows three values for the variable: -1 means never
forward and never change the value of the variable, 0 means don’t forward by default but set this
variable to 1 when two or more interfaces are up, and 1 means always forward. Solaris 2.x changes the
three values to be 0 (never forward), 1 (always forward), and 2 (only forward when two or more

120

interfaces are up).

Older 4.2BSD hosts forwarded datagrams by default, which caused lots of problems for systems
configured improperly. That’s why this kernel option must always default to "never forward" unless
the system administrator specifically enables forwarding.

9.5 ICMP Redirect Errors
The ICMP redirect error is sent by a router to the sender of an IP datagram when the datagram should
have been sent to a different router. The concept is simple, as we show in the three steps in Figure 9.3.
The only time we’ll see an ICMP redirect is when the host has a choice of routers to send the packet
to. (Recall the earlier example of this we saw in Figure 7.6.)

Figure 9.3 Example of an ICMP redirect.

1. We assume that the host sends an IP datagram to Rl. This routing decision is often made because
Rl is the default router for the host.
2. Rl receives the datagram and performs a lookup in its routing table and determines that R2 is the
correct next-hop router to send the datagram to. When it sends the datagram to R2, Rl detects that
it is sending it out the same interface on which the datagram arrived (the LAN to which the host
and the two routers are attached). This is the clue to a router that a redirect can be sent to the
original sender.
3. Rl sends an ICMP redirect to the host, telling it to send future datagrams to that destination to R2,
instead of Rl.

A common use for redirects is to let a host with minimal routing knowledge build up a better routing
table over time. The host can start with only a default route (either Rl or R2 from our example in
Figure 9.3) and anytime this default turns out to be wrong, it’ll be informed by that default router with
a redirect, allowing the host to update its routing table accordingly. ICMP redirects allow TCP/IP hosts
to be dumb when it comes to routing, with all the intelligence in the routers. Obviously Rl and R2 in
our example have to know more about the topology of the attached networks, but all the hosts attached
to the LAN can start with a default route and learn more as they receive redirects.

An Example
We can see ICMP redirects in action on our network (inside front cover). Although we show only
three hosts (aix, solaris, and gemini) and two routers (gateway and netb) on the top
network, there are more than 150 hosts and 10 other routers on this network. Most of the hosts specify
gateway as the default router, since it provides access to the Internet.

How is the author’s subnet (the bottom four hosts in the figure) accessed from the hosts on the
140.252.1 subnet? First recall that if only a single host is at the end of the SLIP link, proxy ARP is
used (Section 4.6). This means nothing special is required for hosts on the top network (140.252.1) to
access the host sun (140.252.1.29). The proxy ARP software in netb handles this.

121

When a network is at the other end of the SLIP link, however, routing becomes involved. One solution
is for every host and router to know that the router netb is the gateway for the network 140.252.13.
This could be done by either a static route in each host’s routing table, or by running a routing daemon
in each host. A simpler way (and the method actually used) is to utilize ICMP redirects.

Let’s run the ping program from the host solaris on the top network to the host bsdi
(140.252.13.35) on the bottom network. Since the subnet IDs are different, proxy ARP can’t be used.
Assuming a static route has not been installed, the first packet sent will use the default route to the
router gateway. Here is the routing table before we run ping:

solaris % netstat -rn
Routing Table:

Destination Gateway Flags Ref Use Interface
127.0.0.1 127.0.0.1 UH 0 848
140.252.1.0 140.252.1.32 U 3 15042 lo0
224.0.0.0 140.252.1.32 U 3 0 le0
default 140.252.1.4 UG 0 5747 le0

(The entry for 224.0.0.0 is for IP multicasting. We describe it in Chapter 12.) If we specify the -v
option to ping, we’ll see any ICMP messages received by the host. We need to specify this to see the
redirect message that’s sent.

solaris % ping -sv bsdi
PING bsdi: 56 data bytes
ICMP Host redirect from gateway gateway (140.252.1.4)
to netb (140.252.1.183) for bsdi (140.252.13.35)
64 bytes from bsdi (140.252.13.35): icmp_seq=0. time=383. Ms
64 bytes from bsdi (140.252.13.35): icmp_seq=l. time=364. Ms
64 bytes from bsdi (140.252.13.35): icmp_seq=2. time=353. Ms
--bsdi PING Statistics--

Before we receive the first ping response, the host receives an ICMP redirect from the default router
gateway. If we then look at the routing table, we’ll see that the new route to the host bsdi has been
inserted. (This new entry is shown in a bolder font.)

Solaris % netstat -rn
Routing Table:

Destination Gateway Flags Ref Use Interface
127.0.0.1 127.0.0.1 UH 0 848
140.252.13.35 140.252.1.183 HGHD 0 2 lo0
140.252.1.0 140.252.1.32 U 3 15045 le0
224.0.0.0 140.252.1.32 U 3 0 le0
default 140.252.1.4 UG 0 5747

122

This is the first time we’ve seen the D flag, which means the route was installed by an ICMP redirect.
The G flag means it’s an indirect route to a gateway (netb), and the H flag means it’s a host route (as
we expect), not a network route.

Since this is a host route, added by a host redirect, it handles only the host bsdi. If we then access the
host svr4, another redirect is generated, creating another host route. Similarly, accessing the host
slip creates another host route. The point here is that each redirect is for a single host, causing a host
route to be added. All three hosts on the author’s subnet (bsdi, svr4, and slip) could also be
handled by a single network route pointing to the router sun. But ICMP redirects create host routes,
not network routes, because the router generating the redirect in this example (gateway) has no
knowledge of the subnet structure on the 140.252.13 network.

More Details
Figure 9.4 shows the format of the ICMP redirect message.

Figure 9.4 ICMP redirect message.

There are four different redirect messages, with different code values, as shown in Figure 9.5.

code Description
0 redirect for network
1 redirect for host
2 redirect for type-of-service and network
3 redirect for type-of-service and host

Figure 9.5 Different code values for ICMP redirect.

There are three IP addresses that the receiver of an ICMP redirect must look at: (1) the IP address that
caused the redirect (which is in the IP header returned as the data portion of the ICMP redirect), (2) the
IP address of the router that sent the redirect (which is the source IP address of the IP datagram
containing the redirect), and (3) the IP address of the router that should be used (which is in bytes 4-7
of the ICMP message).

There are numerous rules about ICMP redirects. First, redirects are generated only by routers, not by
hosts. Also, redirects are intended to be used by hosts, not routers. It is assumed that routers participate
in a routing protocol with other routers, and the routing protocol should obviate the need for redirects.
(This means that in Figure 9.1 the routing table should be updated by either a routing daemon or
redirects, but not by both.)

4.4BSD, when acting as a router, performs the following checks, all of which must be true before an
ICMP redirect is generated.

123

1. The outgoing interface must equal the incoming interface.
2. The route being used for the outgoing datagram must not have been created or modified by an
ICMP redirect, and must not be the router’s default route.
3. The datagram must not be source routed.
4. The kernel must be configured to send redirects.

The kernel variable is named ip_sendredirects, or something similar. (See Appendix E.) Most current systems
(4.4BSD, SunOS 4.1.x, Solaris 2.x, and AIX 3.2.2, for example) enable this variable by default. Other systems such
as SVR4 disable it by default.

Additionally, a 4.4BSD host that receives an ICMP redirect performs some checks before modifying
its routing table. These are to prevent a misbehaving router or host, or a malicious user, from
incorrectly modifying a system’s routing table.

1. The new router must be on a directly connected network.
2. The redirect must be from the current router for that destination.
3. The redirect cannot tell the host to use itself as the router.
4. The route that’s being modified must be an indirect route.

Our final point about redirects is that routers should send only host redirects (codes 1 or 3 from Figure
9.5) and not network redirects. Subnetting makes it hard to specify exactly when a network redirect
can be sent instead of a host redirect. Some hosts treat a received network redirect as a host redirect, in
case a router sends the wrong type.

9.6 ICMP Router Discovery Messages
We mentioned earlier in this chapter that one way to initialize a routing table is with static routes
specified in configuration files. This is often used to set a default entry. A newer way is to use the
ICMP router advertisement and solicitation messages.

The general concept is that after bootstrapping, a host broadcasts or multicasts a router solicitation
message. One or more routers respond with a router advertisement message. Additionally, the routers
periodically broadcast or multicast their router advertisements, allowing any hosts that are listening to
update their routing table accordingly.

RFC 1256 [Deering 1991] specifies the format of these two ICMP messages. Figure 9.6 shows the
format of the ICMP router solicitation message. Figure 9.7 shows the format of the ICMP router
advertisement message sent by routers.

Figure 9.6 Format of ICMP router solicitation message.

124

Figure 9.7 Format of ICMP router advertisement message.

Multiple addresses can be advertised by a router in a single message. Number of addresses is the
number. Address entry size is the number of 32-bit words for each router address, and is always 2.
Lifetime is the number of seconds that the advertised addresses can be considered valid.

One or more pairs of an IP address and a preference then follow. The IP address must be one of the
sending router’s IP addresses. The preference level is a signed 32-bit integer indicating the preference
of this address as a default router address, relative to other router addresses on the same subnet. Larger
values imply more preferable addresses. The preference level 0x80000000 means the corresponding
address, although advertised, is not to be used by the receiver as a default router address. The default
value of the preference is normally 0.

Router Operation
When a router starts up it transmits periodic advertisements on all interfaces capable of broadcasting
or multicasting. These advertisements are not exactly periodic, but are randomized, to reduce the
probability of synchronization with other routers on the same subnet. The normal time interval
between advertisements is between 450 and 600 seconds. The default lifetime for a given
advertisement is 30 minutes.

Another use of the lifetime field occurs when an interface on a router is disabled. In that case the
router can transmit a final advertisement on the interface with the lifetime set to 0.

In addition to the periodic, unsolicited advertisements, a router also listens for solicitations from hosts.
It responds to these solicitations with a router advertisement.

If there are multiple routers on a given subnet, it is up to the system administrator to configure the
preference level for each router as appropriate. For example, the primary default router would have a
higher preference than a backup.

Host Operation
Upon bootstrap a host normally transmits three router solicitations, 3 seconds apart. As soon as a valid
advertisement is received, the solicitations stop.

A host also listens for advertisements from adjacent routers. These advertisements can cause the host’s
default router to change. Also, if an advertisement is not received for the current default, that default
can time out.

As long as the normal default router stays up, that router will send advertisements every 10 minutes,
with a lifetime of 30 minutes. This means the host’s default entry won’t time out, even if one or two
advertisements are lost.

125

Implementation
The router discovery messages are normally generated by and processed by a user process (a daemon).
This adds yet another program updating the routing table in Figure 9.1, although it would only add or
delete a default entry. The daemon would have to be configured to act as a router or a host.

These two ICMP messages are new and not supported by all systems. Solaris 2.x is the only system in our network that
supports these messages (the in.rdisc.daemon). Although the RFC recommends using IP multicasting whenever
possible, router discovery can work using broadcast messages also.

9.7 Summary
The operation of IP routing is fundamental to a system running TCP/IP, be it a host or router. The
routing table entries are simple: up to 5 flag bits, a destination IP address (host, network, or default), a
next-hop router IP address (for an indirect route) or a local interface IP address (for a direct route), and
a pointer to a local interface to use. Host entries have priority over network entries, which have
priority over default entries.

A search of this routing table is made for every IP datagram that the system generates or forwards, and
can be updated by either a routing daemon or ICMP redirects. By default a system should never
forward a datagram unless it has specifically been configured to do so. Static routes can be entered
using the route command, and the newer ICMP router discovery messages can be used to initialize and
dynamically update default entries. Hosts can start with a simple routing table that is updated
dynamically by ICMP redirects from its default router.

Our discussion in this chapter has focused on how a single system uses its routing table. In the next
chapter we examine how routers exchange routing information with each other.

Exercises
9.1 Why do you think both types of ICMP redirects-network and host-exist?

9.2 In the routing table for svr4 shown at the beginning of Section 9.2, is a specific route to the host
slip (140.252.13.65) necessary? What would change if this entry were removed from the routing
table?

9.3 Consider a cable with both 4.2BSD hosts and 4.3BSD hosts. Assume the network ID is 140.1. The
4.2BSD hosts only recognize a host ID of all zero bits as the broadcast address (140.1.0.0), while the
4.3BSD hosts normally send a broadcast using a host ID of all one bits (140.1.255.255). Also, the
4.2BSD hosts by default will try to forward incoming datagrams, even if they have only a single
interface.

Describe the events that happen when the 4.2BSD hosts receives an IP datagram with the destination
address of 140.1.255.255.

9.4 Continue the previous exercise, assuming someone corrects this problem by adding an entry to the
ARP cache on one system on the 140.1 subnet (using the arp command) saying that the IP address
140.1.255.255 has a corresponding Ethernet address of all one bits (the Ethernet broadcast). Describe
what happens now.

126

9.5 Examine your system’s routing table and describe each entry.

127

Dynamic Routing Protocols
10.1 Introduction
Our discussion in the previous chapter dealt with static routing. The routing table entries were created
by default when an interface was configured (for directly connected interfaces), added by the route
command (normally from a system bootstrap file), or created by an ICMP redirect (usually when the
wrong default was used).

This is fine if the network is small, there is a single connection point to other networks, and there are
no redundant routes (where a backup route can be used if a primary route fails). If any of these three
conditions is false, dynamic routing is normally used.

This chapter looks at the dynamic routing protocols used by routers to communicate with each other.
We concentrate on RIP, the Routing Information Protocol, a widely used protocol that is provided with
almost every TCP/IP implementation. We then look at two newer routing protocols, OSPF and BGP.
The chapter finishes with an examination of a new routing technique, called classless interdomain
routing, that is starting to be implemented across the Internet to conserve class B network numbers.

10.2 Dynamic Routing
Dynamic routing occurs when routers talk to adjacent routers, informing each other of what networks
each router is currently connected to. The routers must communicate using a routing protocol, of
which there are many to choose from. The process on the router that is running the routing protocol,
communicating with its neighbor routers, is usually called a routing daemon. As shown in Figure 9.1,
the routing daemon updates the kernel’s routing table with information it receives from neighbor
routers.

The use of dynamic routing does not change the way the kernel performs routing at the IP layer, as we
described in Section 9.2. We called this the routing mechanism. "The kernel still searches its routing
table in the same way, looking for host routes, network routes, and default routes. What changes is the
information placed into the routing table-instead of coming from route commands in bootstrap files,
the routes are added and deleted dynamically by a routing daemon, as routes change over time.

As we mentioned earlier, the routing daemon adds a routing policy to the system, choosing which
routes to place into the kernel’s routing table. If the daemon finds multiple routes to a destination, the
daemon chooses (somehow) which route is best, and which one to insert into the kernel’s table. If the
daemon finds that a link has gone down (perhaps a router crashed or a phone line is out of order), it
can delete the affected routes or add alternate routes that bypass the problem.

In a system such as the Internet, many different routing protocols are currently used. The Internet is
organized into a collection of autonomous systems (ASs), each of which is normally administered by a
single entity. A corporation or university campus often defines an autonomous system. The NSFNET
backbone of the Internet forms an autonomous system, because all the routers in the backbone are
under a single administrative control.

Each autonomous system can select its own routing protocol to communicate between the routers in
that autonomous system. This is called an interior gateway protocol (IGP) or intradomain routing
protocol. The most popular IGP has been the Routing Information Protocol (RIP). A newer IGP is the
Open Shortest Path First protocol (OSPF). It is intended as a replacement for RIP. An older IGP that

128

has fallen out of use is HELLO-the IGP used on the original NSFNET backbone in 1986.

The new Router Requirements RFC [Almquist 1993] states that a router that implements any dynamic routing protocol
must support both OSPF and RIP, and may support other IGPs.

Separate routing protocols called exterior gateway protocols (EGPs) or interdomain routing protocols
are used between the routers in different autonomous systems. Historically (and confusingly) the
predominant EGP has been a protocol of the same name: EGP A newer EGP is the Border Gateway
Protocol (BGP) that is currently used between the NSFNET backbone and some of the regional
networks that attach to the backbone. BGP is intended to replace EGP.

10.3 Unix Routing Daemons
Unix systems often run the routing daemon named routed. It is provided with almost every
implementation of TCP/IP This program communicates using only RIP, which we describe in the next
section. It is intended for small to medium-size networks.

An alternative program is gated. It supports both IGPs and EGPs. [Fedor 1988] describes the early
development of gated. Figure 10.1 compares the various routing protocols supported by routed
and two different versions of gated. Most systems that run a routing daemon run routed, unless they
need support for the other protocols supported by gated.

Daemon Interior Gateway Protocol Exterior Gateway Protocol
HELLO RIP OSPF EGP BGP
routed V1
gated, Version 2 * V1 * V1
gated, Version 3 * V1, V2 V2 * V2, V3

Figure 10.1 Routing protocols supported by routed and gated.

We describe RIP Version I in the next section, the differences with RIP Version 2 in Section 10.5,
OSPF in Section 10.6, and BGP in Section 10.7.

10.4 RIP: Routing Information Protocol
This section provides an overview of RIP, because it is the most widely used (and most often
maligned) routing protocol. The official specification for RIP is RFC 1058 [Hedrick 1988a], but this
RFC was written years after the protocol was widely implemented.

Message Format
RIP messages are carried in UDP datagrams, as shown in Figure 10.2. (We talk more about UDP in
Chapter 11.)

129

Figure 10.2 RIP message encapsulated within a UDP datagram.

Figure 10.3 shows the format of the RIP message, when used with IP addresses.

A command of 1 is a request, and 2 is a reply. There are two other obsolete commands (3 and 4), and
two undocumented ones: poll (5) and poll-entry (6). A request asks the other system to send all or part
of its routing table. A reply contains all or part of the sender’s routing table.

The version is normally 1, although RIP Version 2 (Section 10.5) sets this to 2.

The next 20 bytes specify the address family (which is always 2 for IP addresses), an IP address, and
an associated metric. We’ll see later in this section that RIP metrics are hop counts.

Up to 25 routes can be advertised in a RIP message using this 20-byte format. The limit of 25 is to
keep the total size of the RIP message, 20 x 25+4 = 504, less than 512 bytes. With this limit of 25
routes per message, multiple messages are often required to send an entire routing table.

Figure 10.3 Format of a RIP message.

Normal Operation
Let’s look at the normal operation of routed, using RIP. The well-known port number for RIP is
UDP port 520.

Initialization. When the daemon starts it determines all the interfaces that are up and sends a
request packet out each interface, asking for the other router’s complete routing table. On a
point-to-point link this request is sent to the other end. The request is broadcast if the network
supports it. The destination UDP port is 520 (the routing daemon on the other router).
This request packet has a command of 1 but the address family is set to 0 and the metric is set to
16. This is a special request that asks for a complete routing table from the other end.
Request received. If the request is the special case we just mentioned, then the entire routing table
is sent to the requestor. Otherwise each entry in the request is processed: if we have a route to the
specified address, set the metric to our value, else set the metric to 16. (A metric of 16 is a special
value called "infinity" and means we don’t have a route to that destination.) The response is
returned.
Response received. The response is validated and may update the routing table. New entries can
be added, existing entries can be modified, or existing entries can be deleted.
Regular routing updates. Every 30 seconds, all or part of the router’s entire routing table is sent to
every neighbor router. The routing table is either broadcast (e.g., on an Ethernet) or sent to the

130

other end of a point-to-point link.
Triggered updates. These occur whenever the metric for a route changes. The entire routing table
need not be sent - only those entries that have changed must be transmitted.

Each route has a timeout associated with it. If a system running RIP finds a route that has not been
updated for 3 minutes, that route’s metric is set to infinity (16) and marked for deletion. This means
we have missed six of the 30-second updates from the router that advertised that route. The deletion of
the route from the local routing table is delayed for another 60 seconds to ensure the invalidation is
propagated.

Metrics
The metrics used by RIP are hop counts. The hop count for all directly connected interfaces is 1.
Consider the routers and networks shown in Figure 10.4. The four dashed lines we show are broadcast
RIP messages.

Figure 10.4 Example routers and networks.

Router Rl advertises a route to N2 with a hop count of 1 by sending a broadcast on Nl. (It makes no
sense to advertise a route to Nl in the broadcast sent on Nl.) It also advertises a route to Nl with a hop
count of 1 by sending a broadcast on N2. Similarly, R2 advertises a route to N2 with a metric of 1, and
a route to N3 with a metric of 1.

If an adjacent router advertises a route to another network with a hop count of 1, then our metric for
that network is 2, since we have to send a packet to that router to get to the network. In our example,
the metric to Nl for R2 is 2, as is the metric to N3 for Rl.

As each router sends its routing tables to its neighbors, a route can be determined to each network
within the AS. If there are multiple paths within the AS from a router to a network, the router selects
the path with the smallest hop count and ignores the other paths.

The hop count is limited to 15, meaning RIP can be used only within an AS where the maximum
number of hops between hosts is 15. The special metric of 16 indicates that no route exists to the IP
address.

Problems
As simple as this sounds, there are pitfalls. First, RIP has no knowledge of subnet addressing. If the
normal 16-bit host ID of a class B address is nonzero, for example, RIP can’t tell if the nonzero
portion is a subnet ID or if the IP address is a complete host address. Some implementations use the
subnet mask of the interface through which the RIP information arrived, which isn’t always correct.

Next, RIP takes a long time to stabilize after the failure of a router or a link. The time is usually
measured in minutes. During this settling time routing loops can occur. There are many subtle details
in the implementation of RIP that must be followed to help prevent routing loops and to speed
convergence. RFC 1058 [Hedrick 1988a] contains many details on how RIP should be implemented.

131

The use of the hop count as the routing metric omits other variables that should be taken into
consideration. Also, a maximum of 15 for the metric limits the sizes of networks on which RIP can be
used.

Example
We’ll use the program ripquery, which is available from the gated distribution, to query some
routers for their routing table, ripquery tries to send one of the undocumented requests (named
"poll," a command of 5 from Figure 10.3) to the router, asking for its entire routing table. If no
response is received in 5 seconds, the standard RIP request is issued (command of 1). (Earlier we said
a request with the family set to 0 and the metric set to 16 asks the other router for its entire routing
table.)

Figure 10.5 shows the two routers that we’ll query for their routing table from the host sun. If we
execute ripquery from sun, fetching the routing information from its next-hop router, netb, we get
the following:

sun % ripquery -n netb first message contains 504 bytes
504 bytes from netb lots of other lines deleted
(140.252.1.183): the top Ethernet in Figure 10.5
140.252.1.0, metric 1 the bottom Ethernet in Figure 10.5
140.252.13.0, metric 1 second message with remaining 244
244 bytes from netb bytes
(140.252.1.183): lots of other lines deleted

As we expect, the metric for our subnet that is announced by netb is 1. Additionally, the top Ethernet
that netb is also attached to (140.252.1.0) has a metric of 1. (The -n flag says to print the IP
addresses numerically instead of trying to look up the names.) In this example netb has been
configured to consider all the hosts on the subnet 140.252.13 as directly connected to it - that is, netb
knows nothing about which hosts are actually on the 140.252.13 subnet. Since there is only one
connection point to the 140.252.13 subnet, advertising different metrics for each host makes little
practical sense.

Figure 10.5 Two routers netb and gateway that we’ll query for their routing tables.

Figure 10.6 shows the packet exchange using tcpdump. We specify the SLIP interface with the -i
sl0 option.

132

sun % tcpdump s600 i sl0
1 0.0 sun.2879 > netb.route: rip-poll 24
2 5.014702 (5.0147) sun.2879 > netb.route: rip-req 24
3 5.560427 (0.5457) netb.route > sun.2879: rip-resp 25:
4 5.710251 (0.1498) netb.route > sun.2879: rip-resp 12:

Figure 10.6 tcpdump output while running ripquery program.

The first request issued is the RIP poll command (line 1). This times out after 5 seconds and a normal
RIP request is issued (line 2). The number 24 at the end of lines 1 and 2 is the size of the request
packets in bytes: the 4-byte RIP header (with the command and version) followed by a single 20-byte
address and metric.

Line 3 is the first reply message. The number 25 at the end indicates that 25 address and metric pairs
are in the message, which we calculated earlier to be 504 bytes. This is what ripquery printed
above. We specified the -s600 option to tcpdump telling it to read 600 bytes from the network. This
allows it to receive the entire UDP datagram (instead of just the first portion of it) and it then prints the
contents of the RIP response. We’ve omitted that output.

Line 4 is the second response message from the router, with the next 12 address and metric pairs. We
can calculate the size of this message to be 12 x 20 + 4 = 244, which is what ripquery printed
earlier.

If we go one router beyond netb, to gateway, we expect the metric to our subnet (140.252.13.0) to be
2. We can check this by executing:

sun % ripquery -n gateway
lots of other lines deleted
504 bytes from gateway (140.252.1.4):
the top Ethernet in Figure 10.5
140.252.1.0, metric 1
the bottom Ethernet in Figure 10.5
10.5 140.252.13.0, metric 2

Here the metric for the top Ethernet in Figure 10.5 (140.252.1.0) stays at 1, since that Ethernet is
directly connected to both gateway and netb. Our subnet 140.252.13.0, however, now has the
expected metric of 2.

Another Example
We’ll now watch all the unsolicited RIP updates on an Ethernet and see just what RIP sends on a
regular basis to its neighbors. Figure 10.7 shows the arrangement of many of the noao.edu
networks. We have named the routers Rn for simplicity, where n is the subnet number, except for the
ones we use elsewhere in the text. We show the point-to-point links with dashed lines and the IP
address at each end of these links.

133

Figure 10.7 Many of the noao.edu 140.252 networks.

We’ll run the Solaris 2.x program snoop, which is similar to tcpdump, on the host solaris. We
can run this program without superuser privileges, but only to capture broadcast packets, multicast
packets, or packets sent to the host. Figure 10.8 shows the packets captured during a 60-second period.
We have replaced most of the official host-names with our notation Rn.

solaris % snoop -P -tr udp port 520

0.00000 R6.tuc.noao.edu -> 140.252.1.255 RIP R (1 destinations)
17.02187 gateway.tuc.noao.edu -> 140.252.1.255 RIP R (15
destinations)
20.68009 R10.tuc.noao.edu -> BROADCAST RIP R (4 destinations)

41.34565 R7.tuc.noao.edu ~> 140.252.1.255 RIP R (1 destinations)
47.01870 gateway.tuc.noao.edu -> 140.252.1.255 RIP R (15
destinations)
50.66453 R10.tuc.noao.edu -> BROADCAST RIP R (4 destinations)

Figure 10.8 RIP broadcasts captured at solaris over a 60-second period.

The -P flag captures packets in nonpromiscuous mode, -tr prints the relative time-stamps, and udp
port 520 captures only UDP datagrams with a source or destination port of 520.

The first six packets, from R6, R4, R2, R7, R8, and R3, each advertise just one network. If we looked
at the packets we would see that R6 advertises a route to 140.252.6.0 with a hop count of 1, R4
advertises a route to 140.252.4.0 with a hop count of 1, and so on.

The router gateway, however, advertises 15 routes. We can run snoop with the -v flag and see the
entire contents of the RIP message. (This flag outputs the entire contents of the entire packet: the
Ethernet header, the IP header, the UDP header, and the RIP message. We’ve deleted everything
except the RIP information.) Figure 10.9 shows the output.

134

Compare these advertised hop counts on the 140.252.1 network with the topology shown in Figure
10.7.

A puzzle in the output in Figure 10.8 is why R10 is advertising four networks when Figure 10.7 shows
only three. If we look at the RIP packet with snoop we see the following advertised routes:

RIP: Address Metric
RIP: 140.251.0.0 16 (not reachable)
RIP: 140.252.9.0 1
RIP: 140.252.10.0 1
RIP: 140.252.11.0 1

The route to the class B network 140.251 is bogus and should not be advertised. (It belongs to another
institution, not noao.edu.)

solaris % snoop -P -v -tr udp port 520 host gateway
many lines deleted
RIP: 0pcode = 2 (route response)
RIP: Version = 1
RIP: Address Metric
RIP: 140.252.101.0 1
RIP: 140.252.104.0 1
RIP: 140.252.51.0 2
RIP: 140.252.81.0 2
RIP: 140.252.105.0 2
RIP: 140.252.106.0 2
RIP: 140.252.52.0 3
RIP: 140.252.53.0 3
RIP: 140.252.54.0 3
RIP: 140.252.55.0 3
RIP: 140.252.58.0 3 3
RIP: 140.252.60.0 3 3
RIP: 140.252.82.0 3 3
RIP: 192.68.189.0 3
RIP: 140.252.57.0 4

Figure 10.9 RIP response from gateway.

The notation "BROADCAST" output by snoop in Figure 10.8 for the RIP packet sent by R10 means
the destination IP address is the limited broadcast address 255.255.255.255 (Section 12.2), instead of
the subnet-directed broadcast address (140.252.1.255) that the other routers use.

135

10.5 RIP Version 2
RFC 1388 [Malkin 1993a] defines newer extensions to RIP, and the result is normally called RIP-2.
These extensions don’t change the protocol, but pass additional information in the fields labeled "must
be zero" in Figure 10.3. RIP and RIP-2 can interoperate if RIP ignores the fields that must be zero.

Figure 10.10 is a redo of that figure, as defined by RIP-2. The version is 2 for RIP-2. The routing
domain is an identifier of the routing daemon to which this packet belongs. In a Unix implementation
this could be the daemon’s process ID. This field allows an administrator to run multiple instances of
RIP on a single router, each operating within one routing domain.

The route tag exists to support exterior gateway protocols. It carries an autonomous system number for
EGP and BGP.

The subnet mask for each entry applies to the corresponding IP address. The next-hop IP address is
where packets to the corresponding destination IP address should be sent. A value of 0 in this field
means packets to the destination should be sent to the system sending the RIP message.

Figure 10.10 Format of a RIP-2 message.

A simple authentication scheme is provided with RIP-2. The first 20-byte entry in a RIP message can
specify an address family of 0xffff, with a route tag value of 2. The remaining 16 bytes of the entry
contain a cleartext password.

Finally, RIP-2 supports multicasting in addition to broadcasting (Chapter 12). This can reduce the load
on hosts that are not listening for RIP-2 messages.

10.6 OSPF: Open Shortest Path First
OSPF is a newer alternative to RIP as an interior gateway protocol. It overcomes all the limitations of
RIP. OSPF Version 2 is described in RFC 1247 [Moy 1991].

OSPF is a link-state protocol, as opposed to RIP, which is a distance-vector protocol. The term
distance-vector means the messages sent by RIP contain a vector of distances (hop counts). Each
router updates its routing table based on the vector of these distances that it receives from its
neighbors.

In a link-state protocol a router does not exchange distances with its neighbors. Instead each router
actively tests the status of its link to each of its neighbors, sends this information to its other neighbors,
which then propagate it throughout the autonomous system. Each router takes this link-state
information and builds a complete routing table.

From a practical perspective, the important difference is that a link-state protocol will always converge
faster than a distance-vector protocol. By converge we mean stabilizing after something changes, such
as a router going down or a link going down. Section 9.3 of [Periman 1992] compares other issues

136

between the two types of routing protocols.

OSPF is different from RIP (and many other routing protocols) in that OSPF uses IP directly. That is,
it does not use UDP or TCP. OSPF has its own value for the protocol field in the IP header (Figure
3.1).

Besides being a link-state protocol instead of a distance-vector protocol, OSPF has many other
features that make it superior to RIP.

1. OSPF can calculate a separate set of routes for each IP type-of-service (Figure 3.2). This means
that for any destination there can be multiple routing table entries, one for each IP
type-of-service.
2. Each interface is assigned a dimensionless cost. This can be assigned based on throughput,
round-trip time, reliability, or whatever. A separate cost can be assigned for each IP
type-of-service.
3. When several equal-cost routes to a destination exist, OSPF distributes traffic equally among the
routes. This is called load balancing.
4. OSPF supports subnets: a subnet mask is associated with each advertised route. This allows a
single IP address of any class to be broken into multiple subnets of various sizes. (We showed an
example of this in Section 3.7 and called it variable-length subnets.) Routes to a host are
advertised with a subnet mask of all one bits. A default route is advertised as an IP address of
0.0.0.0 with a mask of all zero bits.
5. Point-to-point links between routers do not need an IP address at each end. These are called
unnumbered networks. This can save IP addresses - a scarce resource these days!
6. A simple authentication scheme can be used. A cleartext password can be specified, similar to the
RIP-2 scheme (Section 10.5).
7. OSPF uses multicasting (Chapter 12), instead of broadcasting, to reduce the load on systems not
participating in OSPF.

With most router vendors supporting OSPF, it will start replacing RIP in many networks.

10.7 BGP: Border Gateway Protocol
BGP is an exterior gateway protocol for communication between routers in different autonomous
systems. BGP is a replacement for the older EGP that was used on the ARPANET. BGP Version 3 is
defined in RFC 1267 [Lougheed and Rekhter 1991].

RFC 1268 [Rekhter and Gross 1991] describes the use of BGP in the Internet. Much of the following
description comes from these two RFCs. Also, during 1993 BGP Version 4 was under development
(see RFC 1467 [Topolcic 1993]) to support CIDR, which we describe in Section 10.8.

A BGP system exchanges network reachability information with other BGP systems. This information
includes the full path of autonomous systems that traffic must transit to reach these networks. This
information is adequate to construct a graph of AS connectivity. Routing loops can then be pruned
from this graph and routing policy decisions can be enforced.

We first categorize an IP datagram in an AS as either local traffic or transit traffic. Local traffic in an
AS either originates or terminates in that AS. That is, either the source IP address or the destination IP
address identifies a host in that AS. Anything else is called transit traffic. A major goal of BGP usage
in the Internet is to reduce transit traffic. An AS can be categorized as one of the following:

137

1. A stub AS has only a single connection to one other AS. A stub AS carries only local traffic.
2. A multihomed AS has connections to more than one other AS, but refuses to carry transit traffic.
3. A transit AS has connections to more than one other AS and is designed, under certain policy
restrictions, to carry both local and transit traffic.

The overall topology of the Internet is then viewed as an arbitrary interconnection of transit,
multihomed, and stub ASs. Stub and multihomed ASs need not use BGP - they can run EGP to
exchange reachability information with transit ASs.

BGP allows for policy-based routing. Policies are determined by the AS administrator and specified to
BGP in configuration files. Policy decisions are not part of the protocol, but policy specifications
allow a BGP implementation to choose between paths when multiple alternatives exist and to control
the redistribution of information. Routing policies are related to political, security, or economic
considerations.

BGP is different from RIP and OSPF in that BGP uses TCP as its transport protocol. Two systems
running BGP establish a TCP connection between themselves and then exchange the entire BGP
routing table. From that point on, incremental updates are sent as the routing table changes.

BGP is a distance vector protocol, but unlike RIP (which announces hops to a destination), BGP
enumerates the route to each destination (the sequence of AS numbers to the destination). This
removes some of the problems associated with distance-vector protocols. An AS is identified by a
16-bit number.

BGP detects the failure of either the link or the host on the other end of the TCP connection by
sending a keepalive message to its neighbor on a regular basis. The recommended time between these
messages is 30 seconds. This application-level keepalive message is independent of the TCP keepalive
option (Chapter 23).

10.8 CIDR: Classless Interdomain Routing
In Chapter 3 we said there is a shortage of class B addresses, requiring sites with multiple networks to
now obtain multiple class C network IDs, instead of a single class B network ID. Although the
allocation of these class C addresses solves one problem (running out of class B addresses) it
introduces another problem: every class C network requires a routing table entry. Classless
Interdomain Routing (CIDR) is a way to prevent this explosion in the size of the Internet routing
tables. It is also called supernetting and is described in RFC 1518 [Rekhter and Li 1993] and RFC
1519 [Fuller et al. 1993], with a overview in [Ford, Rekhter, and Braun 1993]. CIDR has the Internet
Architecture Board’s blessing [Huitema 1993]. RFC 1467 [Topolcic 1993] summarizes the state of
deployment of CIDR in the Internet.

The basic concept in CIDR is to allocate multiple IP addresses in a way that allows summarization into
a smaller number of routing table entries. For example, if a single site is allocated 16 class C
addresses, and those 16 are allocated so that they can be summarized, then all 16 can be referenced
through a single routing table entry on the Internet. Also, if eight different sites are connected to the
same Internet service provider through the same connection point into the Internet, and if the eight
sites are allocated eight different IP addresses that can be summarized, then only a single routing table
entry need be used on the Internet for all eight sites.

138

Three features are needed to allow this summarization to take place.

1. Multiple IP addresses to be summarized together for routing must share the same high-order bits
of their addresses.
2. The routing tables and routing algorithms must be extended to base their routing decisions on a
32-bit IP address and a 32-bit mask.
3. The routing protocols being used must be extended to carry the 32-bit mask in addition to the
32-bit address. OSPF (Section 10.6) and RIP-2 (Section 10.5) are both capable of carrying the
32-bit mask, as is the proposed BGP Version 4.

As an example, RFC 1466 [Gerich 1993] recommends that new class C addresses in Europe be in the
range 194.0.0.0 through 195.255.255.255. In hexadecimal these addresses are from 0xc2000000
through 0xc3ffffff. This represents 65536 different class C network IDs, but they all share the
same high-order 7 bits. In countries other than Europe a single routing table entry with an IP address
of 0xc2000000 and a 32-bit mask of 0xfe000000 (254.0.0.0) could be used to route all of these
65536 class C network IDs to a single point. Subsequent bits of the class C address (that is, the bits
following 194 or 195) can also be allocated hierarchically, perhaps by country or by service provider,
to allow additional summarization within the European routers using additional bits beyond the 7
high-order bits of the 32-bit mask.

CIDR also uses a technique whereby the best match is always the one with the longest match: the one
with the greatest number of one bits in the 32-bit mask. Continuing the example from the previous
paragraph, perhaps one service provider in Europe needs to use a different entry point router than the
rest of Europe. If that provider has been allocated the block of addresses 194.0.16.0 through
194.0.31.255 (16 class C network IDs), routing table entries for just those networks would have an IP
address of 194.0.16.0 and a mask of 255.255.240.0 (0xfffff000). A datagram being routed to the
address 194.0.22.1 would match both this routing table entry and the one for the rest of the European
class C networks. But since the mask 255.255.240 is "longer" than the mask 254.0.0.0, the routing
table entry with the longer mask is used.

The term "classless" is because routing decisions are now made based on masking operations of the
entire 32-bit IP address. Whether the IP address is class A, B, or C makes no difference.

The initial deployment of CIDR is proposed for new class C addresses. Making just this change will
slow down the growth of the Internet routing tables, but does nothing for all the existing routes. This is
the short-term solution. As a long-term solution, if CIDR were applied to all IP addresses, and if
existing IP addresses were reallocated (and all existing hosts renumbered!) according to continental
boundaries and service providers, [Ford, Rekhter, and Braun 1993] claim that the current routing table
consisting of 10,000 network entries could be reduced to 200 entries.

10.9 Summary
There are two basic types of routing protocols: interior gateway protocols (IGPs), for routers within an
autonomous system, and exterior gateway protocols (EGPs), for routers to communicate with routers
in other autonomous systems.

The most popular IGP is the Routing Information Protocol (RIP) with OSPF being a newer IGP that is
gaining widespread use. A new and popular EGP is the Border Gateway Protocol (BGP). In this
chapter we looked at RIP and the types of messages that it exchanges. RIP Version 2 is a recent
enhancement that supports subnetting and other minor improvements. We also described OSPF, BGP,
and classless interdomain routing (CIDR), a newer technique being deployed to reduce the size of the

139

Internet routing tables.

There are a two other OSI routing protocols that you may encounter. Interdomain Routing Protocol
(IDRP) started out as a version of BGP modified for use with OSI addresses instead of 1P.
Intermediate System to Intermediate System Protocol (IS-IS) is the OSI standard IGP. It is used for
routing CLNP (Connectionless Network Protocol), an OSI protocol similar to IP. IS-IS and OSPF are
similar.

Dynamic routing is still a fertile area of internetworking research. The choice of which routing
protocol to use, and which routing daemon to run, is complex. [Periman 1992] provides many of the
details.

Exercises
10.1 In Figure 10.9 which of the routes came to gateway from the router kpno?

10.2 Assume a router has 30 routes to advertise using RIP, requiring one datagram with 25 routes and
another with the remaining 5. What happens if once an hour the first datagram with 25 routes is lost?

10.3The OSPF packet format has a checksum field, but the RIP packet does not. Why?

10.4 What effect does load balancing, as done by OSPF, have on a transport layer?

10.5 Read RFC 1058 for additional details on the implementation of RIP. In Figure 10.8 each router
advertises only the routes that it provides, and none of the other routes that it learned about through the
other router’s broadcasts on the 140.252.1 network. What is this technique called?

10.6 In Section 3.4 we said there are more than 100 hosts on the 140.252.1 subnet in addition to the
eight routers we show in Figure 10.7. What do these 100 hosts do with the eight broadcasts that arrive
every 30 seconds (Figure 10.8)?

140

UDP: User Datagram Protocol
11.1 Introduction
UDP is a simple, datagram-oriented, transport layer protocol: each output operation by a process
produces exactly one UDP datagram, which causes one IP datagram to be sent. This is different from a
stream-oriented protocol such as TCP where the amount of data written by an application may have
little relationship to what actually gets sent in a single IP datagram. Figure 11.1 shows the
encapsulation of a UDP datagram as an IP datagram.

Figure 11.1 UDP encapsulation.

RFC 768 [Postel 1980] is the official specification of UDP.

UDP provides no reliability: it sends the datagrams that the application writes to the IP layer, but there
is no guarantee that they ever reach their destination. Given this lack of reliability, we are tempted to
think we should avoid UDP and always use a reliable protocol such as TCP. After we describe TCP in
Chapter 17 we’ll return to this topic and see what types of applications can utilize UDP.

The application needs to worry about the size of the resulting IP datagram. If it exceeds the network’s
MTU (Section 2.8), the IP datagram is fragmented. This applies to each network that the datagram
traverses from the source to the destination, not just the first network connected to the sending host.
(We defined this as the path MTU in Section 2.9.) We examine IP fragmentation in Section 11.5.

11.2 UDP Header
Figure 11.2 shows the fields in the UDP header.

Figure 11.2 UDP header.

The port numbers identify the sending process and the receiving process. In Figure 1.8 we showed that
TCP and UDP use the destination port number to demultiplex incoming data from IP. Since IP has
already demultiplexed the incoming IP datagram to either TCP or UDP (based on the protocol value in
the IP header), this means the TCP port numbers are looked at by TCP, and the UDP port numbers by
UDP. The TCP port numbers are independent of the UDP port numbers.

141

Despite this independence, if a well-known service is provided by both TCP and UDP, the port
number is normally chosen to be the same for both transport layers. This is purely for convenience and
is not required by the protocols.

The UDP length field is the length of the UDP header and the UDP data in bytes. The minimum value
for this field is 8 bytes. (Sending a UDP datagram with 0 bytes of data is OK.) This UDP length is
redundant. The IP datagram contains its total length in bytes (Figure 3.1), so the length of the UDP
datagram is this total length minus the length of the IP header (which is specified by the header length
field in Figure 3.1).

11.3 UDP Checksum
The UDP checksum covers the UDP header and the UDP data. Recall that the checksum in the IP
header only covers the IP header-it does not cover any data in the IP datagram. Both UDP and TCP
have checksums in their headers to cover their header and their data. With UDP the checksum is
optional, while with TCP it is mandatory.

Although the basics for calculating the UDP checksum are similar to what we described in Section 3.2
for the IP header checksum (the ones complement sum of 16-bit words), there are differences. First,
the length of the UDP datagram can be an odd number of bytes, while the checksum algorithm adds
16-bit words. The solution is to append a pad byte of 0 to the end, if necessary, just for the checksum
computation. (That is, this possible pad byte is not transmitted.)

Next, both UDP and TCP include a 12-byte pseudo-header with the UDP datagram (or TCP segment)
just for the checksum computation. This pseudo-header includes certain fields from the IP header. The
purpose is to let UDP double-check that the data has arrived at the correct destination (i.e., that IP has
not accepted a datagram that is not addressed to this host, and that IP has not given UDP a datagram
that is for another upper layer). Figure 11.3 shows the pseudo-header along with a UDP datagram.

Figure 11.3 Fields used for computation of UDP checksum.

In this figure we explicitly show a datagram with an odd length, requiring a pad byte for the checksum
computation. Notice that the length of the UDP datagram appears twice in the checksum computation.

If the calculated checksum is 0, it is stored as all one bits (65535), which is equivalent in
ones-complement arithmetic. If the transmitted checksum is 0, it indicates that the sender did not
compute the checksum.

If the sender did compute a checksum and the receiver detects a checksum error, the UDP datagram is
silently discarded. No error message is generated. (This is what happens if an IP header checksum
error is detected by IP.)

This UDP checksum is an end-to-end checksum. It is calculated by the sender, and then verified by the
receiver. It is designed to catch any modification of the UDP header or data anywhere between the
sender and receiver.

142

Despite UDP checksums being optional, they should always be enabled. During the 1980s some
computer vendors turned off UDP checksums by default, to speed up their implementation of Sun’s
Network File System (NFS), which uses UDP. While this might be acceptable on a single LAN, where
the cyclic redundancy check on the data-link frame (e.g., Ethernet or token ring frame) can detect most
corruption of the frame, when the datagrams pass through routers, all bets are off. Believe it or not,
there have been routers with software and hardware bugs that have modified bits in the datagrams
being routed. These errors are undetectable in a UDP datagram if the end-to-end UDP checksum is
disabled. Also realize that some data-link protocols (e.g., SLIP) don’t have any form of data-link
checksum.

The Host Requirements RFC requires that UDP checksums be enabled by default. It also states that an
implementation must verify a received checksum if the sender calculated one (i.e., the received
checksum is nonzero). Many implementations violate this, however, and only verify a received
checksum if outgoing checksums are enabled.

tcpdump Output
It is hard to detect whether a particular system has UDP checksums enabled. It is normally impossible
for an application to obtain the checksum field in a received UDP header. To get around this, the
author added another option to the tcpdump program that prints the received UDP checksum. If this
printed value is 0, it means the sending host did not calculate the checksum.

Figure 11.4 shows the output to and from three different systems on our test network (see the figure on
the inside front cover). We ran our sock program (Appendix C), sending a single UDP datagram with
9 bytes of data to the standard echo server.

sun.1900 > gemini.echo: udp 9 (UDP
0.0
1 cksum=6e90)
0.303755 (
2 gemini.echo > sun.1900: udp 9 (UDP
0.3038)
cksum=0)
17.392480 sun.1904 > aix.echo: udp 9 (UDP
3 (17.0887) cksum=6e3b)
4 17.614371 ( aix.echo > sun.1904: udp 9 (UDP
0.2219) cksum=6e3b)
32.092454 sun.1907 > solaris.echo: udp 9 (UDP
5 (14.4781) cksum=6e74)
6 32.314378 ( solaris.echo > sun.1907: udp 9 (UDP
0.2219) cksum=6e74)

Figure 11.4 tcpdump output to see whether other hosts enable UDP checksum.

We can see from this that two of the three systems have UDP checksums enabled.

Also notice that for this simple example the outgoing datagram has the same checksum as the
incoming datagram (lines 3 and 4, 5 and 6). Looking at Figure 11.3 we see that the two IP addresses
are swapped, as are the two port numbers. The other fields in the pseudo-header and the UDP header
are the same, as is the data being echoed. This reiterates that the UDP checksums (indeed, all the
checksums in the TCP/IP protocol suite) are simple 16-bit sums. They cannot detect an error that
swaps two of the 16-bit values.

143

The author also directed a DNS query at each of the eight root name servers described in Section 14.2.
The DNS uses UDP primarily, and only two of the eight had UDP checksums enabled!

Some Statistics
[Mogul 1992] provides counts of various checksum errors on a busy NFS (Network File System)
server that had been up for 40 days. Figure 11.5 summarizes these numbers.

Layer Number of checksum errors Approximate total number of packets
Ethernet 446 170,000,000
IP 14 170,000,000
UDP 5 140,000,000
TCP 350 30,000,000

Figure 11.5 Counts of corrupted packets detected by various checksums.

The final column is only the approximate total for each row, since other protocols are in use at the
Ethernet and IP layers. For example, not all the Ethernet frames are IP datagrams, since minimally
ARP is also used on an Ethernet. Not all IP datagrams are UDP or TCP, since ICMP also uses IP.

Note the much higher percentage of TCP checksum errors compared to UDP checksum errors. This is
probably because the TCP connections on this system tended to be "long distance" (traversing many
routers, bridges, etc.) while the UDP traffic was local. The bottom line is not to trust the data-link
(e.g., Ethernet, token ring, etc.) CRC completely. You should enable the end-to-end checksums all the
time. Also, if your data is valuable, you might not want to trust either the UDP or the TCP checksum
completely, since these are simple checksums and were not meant to catch all possible errors.

We’ll use our sock program to generate some UDP datagrams that we can watch with tcpdump:

bsdi % sock -v -u -i -n4 svr4 discard
connected on 140.252.13.35.1108 to 140.252.13.34.9
bsdi % sock -v -u -i -n4 -w0 svr4 discard
connected on 140.252.13.35.1110 to 140.252.13.34.9

The first time we execute the program we specify the verbose mode (-v) to see the ephemeral port
numbers, specify UDP (-u) instead of the default TCP, and use the source mode (-i) to send data
instead of trying to read and write standard input and output. The -n4 option says to output 4
datagrams (instead of the default 1024) and the destination host is svr4. We described the discard
service in Section 1.12. We use the default output size of 1024 bytes per write.

The second time we run the program we specify -w0, causing 0-length datagrams to be written. Figure
11.6 shows the tcpdump output for both commands.

144

1 0.0 bsdi.1108 > svr4.discard: udp 1024
2 0.002424 ( 0.0024) bsdi.1108 > svr4.discard: udp 1024
5 41.720114 (41.7098) bsdi.1110 > svr4.discard: udp 0

Figure 11.6 tcpdump output when UDP datagrams are sent in one direction.

This output shows the four 1024-byte datagrams, followed by the four 0-length data-grants. Each
datagram followed the previous by a few milliseconds. (It took 41 seconds to type in the second
command.)

There is no communication between the sender and receiver before the first datagram is sent. (We’ll
see in Chapter 17 that TCP must establish a connection with the other end before the first byte of data
can be sent.) Also, there are no acknowledgments by the receiver when the data is received. The
sender, in this example, has no idea whether the other end receives the datagrams.

Finally note that the source UDP port number changes each time the program is run. First it is 1108
and then it is 1110. We mentioned in Section 1.9 that the ephemeral port numbers used by clients are
typically in the range 1024 through 5000, as we see here.

11.5 IP Fragmentation
As we described in Section 2.8, the physical network layer normally imposes an upper limit on the size
of the frame that can be transmitted. Whenever the IP layer receives an IP datagram to send, it
determines which local interface the datagram is being sent on (routing), and queries that interface to
obtain its MTU. IP compares the MTU with the datagram size and performs fragmentation, if
necessary. Fragmentation can take place either at the original sending host or at an intermediate router.

When an IP datagram is fragmented, it is not reassembled until it reaches its final destination. (This
handling of reassembly differs from some other networking protocols that require reassembly to take
place at the next hop, not at the final destination.) The IP layer at the destination performs the
reassembly. The goal is to make fragmentation and reassembly transparent to the transport layer (TCP
and UDP), which it is, except for possible performance degradation. It is also possible for the fragment
of a datagram to again be fragmented (possibly more than once). The information maintained in the IP
header for fragmentation and reassembly provides enough information to do this.

Recalling the IP header (Figure 3.1), the following fields are used in fragmentation. The identification
field contains a unique value for each IP datagram that the sender transmits. This number is copied
into each fragment of a particular datagram. (We now see the use for this field.) The flags field uses
one bit as the "more fragments" bit. This bit is turned on for each fragment comprising a datagram
except the final fragment. The fragment offset field contains the offset of this fragment from the
beginning of the original datagram. Also, when a datagram is fragmented the total length field of each
fragment is changed to be the size of that fragment.

145

Finally, one of the bits in the flags field is called the "don’t fragment" bit. If this is turned on, IP will
not fragment the datagram. Instead the datagram is thrown away and an ICMP error ("fragmentation
needed but don’t fragment bit set," Figure 6.3) is sent to the originator. We’ll see an example of this
error in the next section.

When an IP datagram is fragmented, each fragment becomes its own packet, with its own IP header,
and is routed independently of any other packets. This makes it possible for the fragments of a
datagram to arrive at the final destination out of order, but there is enough information in the IP header
to allow the receiver to reassemble the fragments correctly.

Although IP fragmentation looks transparent, there is one feature that makes it less than desirable: if
one fragment is lost the entire datagram must be retransmitted. To understand why this happens,
realize that IP itself has no timeout and retransmission-that is the responsibility of the higher layers.
(TCP performs timeout and retransmission, UDP doesn’t. Some UDP applications perform timeout
and retransmission themselves.) When a fragment is lost that came from a TCP segment, TCP will
time out and retransmit the entire TCP segment, which corresponds to an IP datagram. There is no way
to resend only one fragment of a datagram. Indeed, if the fragmentation was done by an intermediate
router, and not the originating system, there is no way for the originating system to know how the
datagram was fragmented, For this reason alone, fragmentation is often avoided. [Kent and Mogul
1987] provide arguments for avoiding fragmentation.

Using UDP it is easy to generate IP fragmentation. (We’ll see later that TCP tries to avoid
fragmentation and that it is nearly impossible for an application to force TCP to send segments large
enough to require fragmentation.) We can use our sock program and increase the size of the datagram
until fragmentation occurs. On an Ethernet the maximum amount of data in a frame is 1500 bytes
(Figure 2.1), which leaves 1472 bytes for our data, assuming 20 bytes for the IP header and 8 bytes for
the UDP header. We’ll run our sock program, with data sizes of 1471, 1472, 1473, and 1474 bytes. We
expect the last two to cause fragmentation:

bsdi % sock -u -i -nl -wl471 svr4 discard

Figure 11.7 shows the corresponding tcpdump output.

146

1 0.0 bsdi-1112 > svr4.discard: udp 1471
21.008303
2 bsdi.lll4 > svr4.discard: udp 1472
(21.0083)
50.449704 bsdi.lll6 > svr4.discard: udp 1473 (frag
3
(29.4414) 26304:1480@0+)
50.450040 (
4 bsdi > svr4: (frag 26304:l@1480)
0.0003)
75.328650 bsdi.1118 > svr4.discard: udp 1474 (frag
5
(24.8786) 26313:1480@0+)
75.328982 (
6 bsdi > svr4: (frag 26313:2@1480)
0.0003)

Figure 11.7 Watching fragmentation of UDP datagrams.

The first two UDP datagrams (lines 1 and 2) fit into Ethernet frames, and are not fragmented. But the
length of the IP datagram corresponding to the write of 1473 bytes is 1501, which must be fragmented
(lines 3 and 4). Similarly the datagram generated by the write of 1474 bytes is 1502, and is also
fragmented (lines 5 and 6).

When the IP datagram is fragmented, tcpdump prints additional information. First, the output frag
26304 (lines 3 and 4) and frag 26313 (lines 5 and 6) specify the value of the identification field in
the IP header.

The next number in the fragmentation information, the 1480 between the colon and the at sign in line
3, is the size, excluding the IP header. The first fragment of both datagrams contains 1480 bytes of
data: 8 bytes for the UDP header and 1472 bytes of user data. (The 20-byte IP header makes the packet
exactly 1500 bytes.) The second fragment of the first datagram (line 4) contains 1 byte of data-the
remaining byte of user data. The second fragment of the second datagram (line 6) contains the
remaining 2 bytes of user data.

Fragmentation requires that the data portion of the generated fragments (that is, everything excluding
the IP header) be a multiple of 8 bytes for all fragments other than the final one. In this example, 1480
is a multiple of 8.

The number following the at sign is the offset of the data in the fragment, from the start of the
datagram. The first fragment of both datagrams starts at 0 (lines 3 and 5) and the second fragment of
both datagrams starts at byte offset 1480 (lines 4 and 6). The plus sign following this offset that is
printed for the first fragment of both datagrams means there are more fragments comprising this
datagram. This plus sign corresponds to the "more fragments" bit in the 3-bit flags in the IP header.
The purpose of this bit is to let the receiver know when it has completed the reassembly of all the
fragments for a datagram.

Finally, notice that lines 4 and 6 (fragments other than the first) omit the protocol (UDP) and the
source and destination ports. The protocol could be printed, since it’s in the IP header that’s copied
into the fragments. The port numbers, however, are in the UDP header, which only occurs in the first
fragment.

147

Figure 11.8 shows what’s happening with the third datagram that is sent (with 1473 bytes of user
data). It reiterates that any transport layer header appears only in the first fragment.

Also note the terminology: an IP datagram is the unit of end-to-end transmission at the IP layer
(before fragmentation and after reassembly), and a packet is the unit of data passed between the IP
layer and the link layer. A packet can be a complete IP datagram or a fragment of an IP datagram.

Figure 11.8 Example of UDP fragmentation.

11.6 ICMP Unreachable Error (Fragmentation Required)
Another variation of the ICMP unreachable error occurs when a router receives a datagram that
requires fragmentation, but the don’t fragment (DF) flag is turned on in the IP header. This error can
be used by a program that needs to determine the smallest MTU in the path to a destination-called the
path MTU discovery mechanism (Section 2.9).

Figure 11.9 shows the format of the ICMP unreachable error for this case. This differs from Figure
6.10 because bits 16-31 of the second 32-bit word can provide the MTU of the next hop, instead of
being 0.

Figure 11.9 ICMP unreachable error when fragmentation required but don’t fragment bit set.

If a router doesn’t provide this newer format ICMP error, the next-hop MTU is set to 0.

The new Router Requirements RFC [Almquist 1993] states that a router must generate this newer form
when originating this ICMP unreachable error.

Example
A problem encountered by the author involving fragmentation and this ICMP error is trying to
determine the MTU on the dialup SLIP link from the router netb to the host sun. We know the MTU
of this link from sun to netb: it’s part of the SLIP configuration process when SLIP was installed in
the host sun, plus we saw it with the netstat command in Section 3.9. We want to determine the
MTU in the other direction also. (In Chapter 25 we’ll see how to determine this using SNMP.) On a
point-to-point link, it is not required that the MTU be the same in both directions.

The technique used was to run ping on the host solaris, to the host bsdi, increasing the size of
the data packets until fragmentation was seen on the incoming packets. This is shown in Figure 11.10.

148

Figure 11.10 Systems being used to determine MTU of SLIP link from netb to sun.

tcpdump was run on the host sun, watching the SLIP link, to see when fragmentation occurred. No
fragmentation was observed and everything was fine until the size of the data portion of the ping
packet was increased from 500 to 600 bytes. The incoming echo requests were seen (there was still no
fragmentation), but the echo replies disappeared.

To track this down, tcpdump was also run on bsdi, to see what it was receiving and sending. Figure
11.11 shows the output.

1 0.0 solaris > bsdi: icmp: echo request (DF)
2 0.000000 (0.0000) bsdi > solaris: icmp: echo reply (DF)
sun > bsdi: icmp: solaris unreachable -
3 0.000000 (0.0000)
need to frag, mtu = 0 (DF)
4 0.738400 (0.7384) solaris > bsdi: icmp: echo request (DF)
5 0.748800 (0.0104) bsdi > solaris: icmp: echo reply (DF)
sun > bsdi: icmp: solaris unreachable -
6 0.748800 (0.0000)
need to frag, mtu = 0 (DF)

Figure 11.11 tcpdump output for ping of bsdi from solaris with 600-byte IP datagram.

First, the notation (DF) in each line means the don’t fragment bit is turned on in the IP header. It turns
out that Solaris 2.2 normally turns this bit on, as part of its implementation of the path MTU discovery
mechanism.

Line 1 shows that the echo request got through the router netb to sun without being fragmented, and
with the DF bit set, so we know that the SLIP MTU of netb has not been reached yet.

Next, notice in line 2 that the DF flag is copied into the echo reply. This is what causes the problem.
The echo reply is the same size as the echo request (just over 600 bytes), but the MTU on sun’s
outgoing SLIP interface is 552. The echo reply needs to be fragmented, but the DF flag is set. This
causes sun to generate the ICMP unreachable error back to bsdi (where it’s discarded).

This is why we never saw any echo replies on solaris. The replies never got past sun. Figure
11.12 shows the path of the packets.

149

Figure 11.12 Packets exchanged in example.

Finally, the notation mtu=0 in lines 3 and 6 of Figure 11.11 indicates that sun does not return the
MTU of the outgoing interface in the ICMP unreachable message, as shown in Figure 11.9. (In Section
25.9 we return to this problem and use SNMP to determine that the MTU of the SLIP interface on netb
is 1500.)

11.7 Determining the Path MTU Using Traceroute
Although most systems don’t support the path MTU discovery feature, we can easily modify a version
of traceroute (Chapter 8) to let us determine the path MTU. What we’ll do is send packets with
the "don’t fragment" bit set. The size of the first packet we send will equal the MTU of the outgoing
interface, and whenever we receive an ICMP "can’t fragment" error (which we described in the
previous section) we’ll reduce the size of the packet. If the router sending the ICMP error sends the
newer version that includes the MTU of the outgoing interface, we’ll use that value; otherwise we’ll
try the next smallest MTU. As RFC 1191 [Mogul and Deering 1990] states, there are a limited number
of MTUs, so our program has a table of the likely values and moves to the next smallest value.

Let’s first try it from our host sun to the host slip, knowing that the SLIP link has an MTU of 296:

sun % traeeroute.pmtu slip
traceroute to slip (140.252.13.65), 30 hops max
outgoing MTU = 1500
1 bsdi (140.252.13.35) 15 ms 6 ms 6 ms
2 bsdi (140.252.13.35) 6 ms
fragmentation required and DF set, trying new MTU = 1492
2 slip (140.252.13.65) 377 ms 377 ms 377 ms

In this example the router bsdi does not return the MTU of the outgoing interface in the ICMP error,
so we step through the likely values for the MTU. The first line of output for a TTL of 2 prints a
hostname of bsdi, but that’s because it’s the router returning the ICMP error. The final line of output
for a TTL of 2 is what we’re looking for.

It’s not hard to modify the ICMP code on bsdi to return the MTU of the outgoing interface, and if we
do that and rerun our program, we get the following output:

150

sun % traceroute.pmtu slip
traceroute to slip (140.252.13.65), 30 hops max
outgoing MTU = 1500
1 bsdi (140.252.13.35) 53 ms 6 ms 6 ms
2 bsdi (140.252.13.35) 6 ms
fragmentation required and DF set, next hop MTU = 296
2 slip (140.252.13.65) 377 ms 378 ms 377 ms

Here we don’t have to try eight different values for the MTU before finding the right one-the router
returns the correct value.

The Worldwide Internet
As an experiment, this modified version of traceroute was run numerous times to various hosts
around the world. Fifteen countries (including Antarctica) were reached and various transatlantic and
transpacific links were used. Before doing this, however, the MTU of the dialup SLIP link between the
author’s subnet and the router netb (Figure 11.12) was increased to 1500, the same as an Ethernet.

Out of 18 runs, only 2 had a path MTU of less than 1500. One of the transatlantic links had an MTU of
572 (a value not even listed as a likely value in RFC 1191) and the router did return the newer format
ICMP error. Another link, between two routers in Japan, wouldn’t handle a 1500-byte frame, and the
router did not return the newer format ICMP error. Setting the MTU down to 1006 did work.

The conclusion we can make from this experiment is that many, but not all, WANs today can handle
packets larger than 512 bytes. Using the path MTU discovery feature will allow applications to take
advantage of these larger MTUs.

11.8 Path MTU Discovery with UDP
Let’s examine the interaction between an application using UDP and the path MTU discovery
mechanism. We want to see what happens when the application writes datagrams that are too big for
some intermediate link.

Example
Since the only system that we’ve been using that supports the path MTU discovery mechanism is
Solaris 2.x, we’ll use it as the source host to send 650-byte datagrams to slip. Since our host slip
sits behind a SLIP link with an MTU of 296, any UDP datagram greater than 268 bytes (296 -20-8)
with the "don’t fragment" bit set should cause the router bsdi to generate the ICMP "can’t fragment"
error. Figure 11.13 shows the topology and the MTUs.

Figure 11.13 Systems used for path MTU discovery using UDP.

151

The following command generates ten 650-byte UDP datagrams, with a 5-second pause between each
datagram:

solaris % sock -u -i -n10 -w650 -p5 slip discard

Figure 11.14 shows the tcpdump output. When this example was run, the router bsdi was set to not
return the next-hop MTU as part of the ICMP "can’t fragment" error.

The first datagram is sent with the DF bit set (line 1) and generates the expected error from the router
bsdi (line 2). What’s puzzling is that the next datagram is also sent with the DF bit set (line 3) and
generates the same ICMP error (line 4). We would expect this datagram to be sent with the DF bit off.

On line 5 it appears IP has finally learned that datagrams to this destination should not be sent with the
DF bit set, so IP goes ahead and fragments the datagrams at the source host. This is different from
earlier examples where IP sends the datagram that is passed to it by UDP and allows the router with
the smaller MTU (bsdi in this case) to

1 0.0 solaris.38196 > slip.discard: udp 650 (DF)
bsdi > solaris: icmp:
0.004218
2 slip unreachable - need to frag, mtu = 0
(0.0042)
(DF)
4.980528
3 solaris.38196 > slip.discard: udp 650 (DF)
(4.9763)
4.984503
(0.0040)
(DF)
9.870407 solaris.38196 > slip.discard: udp 650 (frag
5
(4.8859) 47942:552@0+)
9.960056
6 solaris > slip: (frag 47942:106(3552)
(0.0896)
14.940338
(4.9803)
14.944466
(0.0041)
(DF)
9
(4.9455) 47944:552@0+)
19.950463
10 solaris > slip: (frag 47944:106@552)
(0.0604)
11
(4.9199) 47945:552@0+)
24.960038
12 solaris > slip: (frag 47945:1060552)
(0.0896)

152

13
(4.9201) 47946:552@0+)
29.940498
(0.0603)
15
(4.9201) 47947:552@0+)
34.950051
(0.0894)
17
(4.9202) 47948:552@0+)
39.930443
(0.0602)
44.940485
(5.0100)
44.944432
(0.0039)
(DF)

Figure 11.14 Path MTU discovery using UDP.

do the fragmentation. Since the ICMP "can’t fragment" message didn’t specify the next-hop MTU, it
appears that IP guesses that an MTU of 576 is OK. The first fragment (line 5) contains 544 bytes of
UDP data, the 8-byte UDP header, and the 20-byte IP header, for a total IP datagram size of 572 bytes.
The second fragment (line 6) contains the remaining 106 bytes of UDP data and a 20-byte IP header.

Unfortunately the next datagram, line 7, has its DF bit set, so it’s discarded by bsdi and the ICMP
error returned. What has happened here is that an IP timer has expired telling IP to see if the path
MTU has increased by setting the DF bit again. We see this happen again on lines 19 and 20.
Comparing the times on lines 7 and 19 it appears that IP turns on the DF bit, to see if the path MTU
has increased, every 30 seconds.

This 30-second timer value is way too small. RFC 1191 recommends a value of 10 minutes. It can be
changed by modifying the parameter ip_ire_pathmtu_interval (Section E.4). Also there is no
way in Solaris 2.2 to turn off this path MTU discovery for a single UDP application or for all UDP
applications. It can only be enabled or disabled on a systemwide basis by changing the parameter
ip_path_mtu_discovery. As we can see from this example, enabling path MTU discovery when
UDP applications write datagrams that will probably be fragmented can cause datagrams to be
discarded.

The maximum datagram size assumed by the IP layer on solaris (576 bytes) is not right. In Figure
11.13 we see that the real MTU is 296 bytes. This means the fragments generated by solaris will
be fragmented again by bsdi. Figure 11.15 shows the tcpdump output collected on the destination
host (slip) for the first datagram that arrives (lines 5 and 6 from Figure 11.14).

153

solaris.38196 > slip.discard: udp 650 (frag
1 0.0
47942:272@0+)
0.304513
2 solaris > slip: (frag 47942:272@272+)
(0.3045)
0.334651
(0.0301)
0.466642
(0.1320)

Figure 11.15 First datagram arriving at host slip from solaris.

In this example the host solaris should not fragment the outgoing datagrams but should turn off the
DF bit and let the router with the smaller MTU do the fragmentation.

Now we’ll run the same example but modify the router bsdi to return the next-hop MTU in the ICMP
"can’t fragment" error. Figure 11.16 shows the first six lines of the tcpdump output.

1 0.0 solaris. 37974 > slip.discard: udp 650 (DF)
0.004199
(0.0042)
(DF)
4.950193
(4.9460)
4.954325
(0.0041)
(DF)
5
(4.8255) 35278:272@0+)
9.930018
(0.1502)
9.990170
(0.0602)

Figure 11.16 Path MTU discovery using UDP.

Again, the first two datagrams are sent with the DF bit set, and both elicit the ICMP error. The ICMP
error now specifies the next-hop MTU of 296.

In lines 5, 6, and 7 we see the source host perform fragmentation, similar to Figure 11.14. But
knowing the next-hop MTU, only three fragments are generated, compared to the four fragments
generated by the router bsdi in Figure 11.15.

154

11.9 Interaction Between UDP and ARP
Using UDP we can see an interesting (and often unmentioned) interaction with UDP and typical
implementations of ARP.

We use our sock program to generate a single UDP datagram with 8192 bytes of data. We expect this
to generate six fragments on an Ethernet (see Exercise 11.3). We also assure that the ARP cache is
empty before running the program, so that an ARP request and reply must be exchanged before the
first fragment is sent.

bsdi % arp -a verify ARP cache is empty
bsdi % sock -u -i -nl -w8192 svr4 discard

We expect the first fragment to cause an ARP request to be sent. Five more fragments are generated by
IP and this presents two timing questions that we’ll need to use tcpdump to answer: are the
remaining fragments ready to be sent before the ARP reply is received, and if so, what does ARP do
with multiple packets to a given destination when it’s waiting for an ARP reply? Figure 11.17 shows
the tcpdump output.

1 0.0 arp who-has svr4 tell bsdi
2 0.001234 (0.0012) arp who-has svr4 tell bsdi
9 0.011127 (0.0012) bsdi > svr4: (frag 10863:800@7400)

Figure 11.17 Packet exchange when an 8192-byte UDP datagram is sent on an Ethernet.

There are a few surprises in this output. First, six ARP requests are generated before the first ARP
reply is returned. What we guess is happening is that IP generates the six fragments rapidly, and each
one causes an ARP request.

155

Next, when the first ARP reply is received (line 7) only the last fragment is sent (line 9)! It appears
that the first five fragments have been discarded. Indeed, this is the normal operation of ARP. Most
implementations keep only the last packet sent to a given destination while waiting for an ARP reply.

The Host Requirements RFC requires an implementation to prevent this type of ARP flooding
(repeatedly sending an ARP request for the same IP address at a high rate). The recommended
maximum rate is one per second. Here we see six ARP requests in 4.3 ms.

The Host Requirements RFC states that ARP should save at least one packet, and this should be the
latest packet. That’s what we see here.

Another unexplained anomaly in this output is that svr4 sends back seven ARP replies, not six.

The final point worth mentioning is that tcpdump was left to run for 5 minutes after the final ARP
reply was returned, waiting to see if svr4 sent back an ICMP "time exceeded during reassembly" error.
The ICMP error was never sent. (We showed the format of this message in Figure 8.2. A code of I
indicates that the time was exceeded during the reassembly of a datagram.)

The IP layer must start a timer when the first fragment of a datagram appears. Here "first" means the
first arrival of any fragment for a given datagram, not the first fragment (with a fragment offset of 0).
A normal timeout value is 30 or 60 seconds. If all the fragments for this datagram have not arrived
when the timer expires, all these fragments are discarded. If this were not done, fragments that never
arrive (as we see in this example) could eventually cause the receiver to run out of buffers.

There are two reasons we don’t see the ICMP message here. First, most Berkeley-derived
implementations never generate this error! These implementations do set a timer, and do discard all
fragments when the timer expires, but the ICMP error is never generated. Second, the first
fragment-the one with an offset of 0 containing the UDP header-was never received. (It was the first of
the five packets discarded by ARP.) An implementation is not required to generate the ICMP error
unless this first fragment has been received. The reason is that the receiver of the ICMP error couldn’t
tell which user process sent the datagram that was discarded, because the transport layer header is not
available. It’s assumed that the upper layer (either TCP or the application using UDP) will eventually
time out and retransmit.

In this section we’ve used IP fragmentation to see this interaction between UDP and ARP. We can also
see this interaction if the sender quickly transmits multiple UDP datagrams. We chose to use
fragmentation because the packets get generated quickly by IP, faster than multiple datagrams can be
generated by a user process.

As unlikely as this example might seem, it occurs regularly. NFS sends UDP datagrams whose length
just exceeds 8192 bytes. On an Ethernet these are fragmented as we’ve indicated, and if the
appropriate ARP cache entry times out, you can see what we’ve shown here. NFS will time out and
retransmit, but the first IP datagram can still be discarded because of ARP’s limited queue.

11.10 Maximum UDP Datagram Size
Theoretically, the maximum size of an IP datagram is 65535 bytes, imposed by the 16-bit total length
field in the IP header (Figure 3.1). With an IP header of 20 bytes and a UDP header of 8 bytes, this
leaves a maximum of 65507 bytes of user data in a UDP datagram. Most implementations, however,
provide less than this maximum.

156

"There are two limits we can encounter. First the application program may be limited by its
programming interface. The sockets API (Section 1.15) provides a function that the application can
call to set the size of the receive buffer and the send buffer. For a UDP socket, this size is directly
related to the maximum size UDP datagram the application can read or write. Most systems today
provide a default of just over 8192 bytes for the maximum size of a UDP datagram that can be read or
written. (This default is because 8192 is the amount of user data that NFS reads and writes by default.)

The next limitation comes from the kernel’s implementation of TCP/IP. There may be implementation
features (or bugs) that limit the size of an IP datagram to less than 65535 bytes.

The author experimented with various UDP datagram sizes, using the sock program. Using the
loopback interface under SunOS 4.1.3, the maximum size IP datagram was 32767 bytes. All higher
values failed. But going across an Ethernet from BSD/386 to SunOS 4.1.3, the maximum size IP
datagram the Sun could accept was 32786 (that is, 32758 bytes of user data). Using the loopback
interface under Solaris 2.2, the maximum 65535-byte IP datagram could be sent and received. From
Solaris 2.2 to AIX 3.2.2, the maximum 65535-byte IP datagram could be transferred. Obviously this
limit depends on the source and destination implementations.

We mentioned in Section 3.2 that a host is required to receive at least a 576-byte IP datagram. Many
UDP applications are designed to restrict their application data to 512 bytes or less, to stay below this
limit. We saw this in Section 10.4, for example, where the Routing Information Protocol always sent
less than 512 bytes of data per datagram. We’ll encounter this same limit with other UDP applications:
the DNS (Chapter 14), TFTP (Chapter 15), BOOTP (Chapter 16), and SNMP (Chapter 25).

Datagram Truncation
Just because IP is capable of sending and receiving a datagram of a given size doesn’t mean the
receiving application is prepared to read that size. UDP programming interfaces allow the application
to specify the maximum number of bytes to return each time. What happens if the received datagram
exceeds the size the application is prepared to deal with?

Unfortunately the answer depends on the programming interface and the implementation.

The traditional Berkeley version of the sockets API truncates the datagram, discarding any excess data.
Whether the application is notified depends on the version. (4.3BSD Reno and later can notify the
application that the datagram was truncated.)

The sockets API under SVR4 (including Solaris 2.x) does not truncate the datagram. Any excess data
is returned in subsequent reads. The application is not notified that multiple reads are being fulfilled
from a single UDP datagram.

The TLI API does not discard the data. Instead a flag is returned indicating that more data is available,
and subsequent reads by the application return the rest of the datagram.

When we discuss TCP we’ll see that it provides a continuous stream of bytes to the application,
without any message boundaries. TCP passes the data to the application in whatever size reads the
application asks for-there is never any data loss across this interface.

157

11.11 ICMP Source Quench Error
Using UDP we are also able to generate the ICMP "source quench" error. This is an error that may be
generated by a system (router or host) when it receives datagrams at a rate that is too fast to be
processed. Note the qualifier "may." A system is not required to send a source quench, even if it runs
out of buffers and throws datagrams away.

Figure 11.18 shows the format of the ICMP source quench error. We have a perfect scenario with our
test network for generating this error. We can send datagrams from bsdi to the router sun across the
Ethernet that must be routed across the dialup SLIP link. Since the SLIP link is about 1000 times
slower than the Ethernet, we should easily be able to overrun its buffer space. The following command
sends 100 1024-byte datagrams from the host bsdi through the router sun to solaris. We send the
datagrams to the standard discard service, where they’ll be ignored:

bsdi % sock -u -i -w1024 -n100 solaris discard

Figure 11.18 ICMP source quench error.

Figure 11.19 shows the tcpdump output corresponding to this command.

1 0.0 bsdi .1403 > solaris.discard: udp 1024
26 lines that we don’t show
27 0.10 (0.00) bsdi.1403 > solaris.discard: udp 1024
28 0.11 (0.01) sun > bsdi: icmp: source quench
29 0.11 (0.00) bsdi.1403 > solaris.discard: udp 1024
142 lines that we don’t show
173 0.71 (0.06) bsdi. 1403 > solaris.discard: udp 1024

Figure 11.19 ICMP source quench from the router sun.

We have removed lots of lines from this output; there is a pattern. The first 26 datagrams are received
without an error; we show the output only for the first. Starting with our 27th datagram, however,
every time we send a datagram, we receive a source quench in return. There are a total of 26 + (74 x 1)
= 174 lines of output.

158

From our serial line throughput calculations in Section 2.10, it takes just over 1 second to transfer a
1024-byte datagram at 9600 bits/sec. (In our example it should take longer than this since the 20 + 8 +
1024 byte datagram will be fragmented because the MTU of the SLIP link from sun to netb is 552
bytes.) But we can see from the timing in Figure 11.19 that the router sun receives all 100 datagrams
in less than 1 second, before the first one is through the SLIP link. It’s not surprising that we used up
many of its buffers.

Although RFC 1009 [Braden and Postel 1987] requires a router to generate source quenches when it
runs out of buffers, the new Router Requirements RFC [Almquist 1993] changes this and says that a
router must not originate source quench errors. The current feeling is to deprecate the source quench
error, since it consumes network bandwidth and is an ineffective and unfair fix for congestion.

Another point to make regarding this example is that our sock program either never received a
notification that the source quenches were being received, or if it did, it appears to have ignored them..
It turns out that BSD implementations normally ignore received source quenches if the protocol is
UDP. (TCP is notified, and slows down the data transfer on the connection that generated the source
quench, as we discuss in Section 21.10.) Part of the problem is that the process that generated the data
that caused the source quench may have already terminated when the source quench is received.
Indeed, if we use the Unix time program to measure how long our sock program takes to run, it only
executes for about 0.5 seconds. But from Figure 11.19 we see that some of the source quenches are
received 0.71 seconds after the first datagram was sent, after the process has terminated. What is
happening is that our program writes 100 datagrams and terminates. But not all 100 datagrams have
been sent-some are queued for output.

This example reiterates that UDP is an unreliable protocol and illustrates the value of end-to-end flow
control. Even though our sock program successfully wrote 100 datagrams to its network, only 26
were really sent to the destination. The other 74 were probably discarded by the intermediate router.
Unless we build some form. of acknowledgment into the application, the sender has no idea whether
the receiver really got the data.

11.12 UDP Server Design
There are some implications in using UDP that affect the design and implementation of a server. The
design and implementation of clients is usually easier than that of servers, which is why we talk about
server design and not client design. Servers typically interact with the operating system and most
servers need a way to handle multiple clients at the same time.

Normally a client starts, immediately communicates with a single server, and is done. Servers, on the
other hand, start and then go to sleep, waiting for a client’s request to arrive. In the case of UDP, the
server wakes up when a client’s datagram arrives, probably containing a request message of some
form from the client.

Our interest here is not in the programming aspects of clients and servers ([Stevens 1990] covers all
those details), but in the protocol features of UDP that affect the design and implementation of a server
using UDP. (We examine the details of TCP server design in Section 18.11.) Although some of the
features we describe depend on the implementation of UDP being used, the features are common to
most implementations.

159

Client IP Address and Port Number
What arrives from the client is a UDP datagram. The IP header contains the source and destination IP
addresses, and the UDP header contains the source and destination UDP port numbers. When an
application receives a UDP datagram, it must be told by the operating system who sent the
message-the source IP address and port number.

This feature allows an iterative UDP server to handle multiple clients. Each reply is sent back to the
client that sent the request.

Destination IP Address
Some applications need to know who the datagram was sent to, that is, the destination IP address. For
example, the Host Requirements RFC states that a TFTP server should ignore received datagrams that
are sent to a broadcast address. (We describe broadcasting in Chapter 12 and TFTP in Chapter 15.)

This requires the operating system to pass the destination IP address from the received UDP datagram
to the application. Unfortunately, not all implementations provide this capability.

The sockets API provides this capability with the IP_RECVDSTADDR socket option. Of the systems
used in the text, only BSD/386, 4.4BSD, and AIX 3.2.2 support this option. SVR4, SunOS 4.x, and
Solaris 2.x don’t support it.

UDP Input Queue
We said in Section 1.8 that most UDP servers are iterative servers. This means a single server process
handles all the client requests on a single UDP port (the server’s well-known port).

Normally there is a limited size input queue associated with each UDP port that a application is using.
This means that requests that arrive at about the same time from different clients are automatically
queued by UDP. The received UDP datagrams are passed to the application (when it asks for the next
one) in the order they were received. It is possible, however, for this queue to overflow, causing the
kernel’s UDP module to discard incoming datagrams. We can see this with the following experiment.
We start our sock program on the host bsdi running as a UDP server:

bsdi % sock -s -u -v -E -R256 -r256 -P30 6666
from 140.252.13.33, to 140.252.13.63: from sun to broad cast
1111111111 address
from 140.252.13.34, to 140.252.13.35: from svr4, to unicast
4444444444444 address

We specify the following flags: -s to run as a server, -u for UDP, -v to print the client’s IP address,
and -E to print the destination IP address (which is supported by this system). Additionally we set the
UDP receive buffer for this port to 256 bytes (-R), along with the size of each application read (-r).
The flag -P30 tells it to pause for 30 seconds after creating the UDP port, before reading the first
datagram. This gives us time to start the clients on two other hosts, send some datagrams, and see how
the receive queueing works.

160

Once the server is started, and is in its 30-second pause, we start one client on the host sun and send
three datagrams:

sun % sock -u -v 140.252.13.63 6666 to Ethernet broadcast address
connected on 140.252.13.33.1252 to 140.252.13.63.6666
1111111111 11 bytes of data (with newline)

The destination address is the broadcast address (140.252.13.63). We also start a second client on the
host svr4 and send another three datagrams:

svr4 % sock -u -v bsdi 6666
connected on 0.0.0.0.1042 to 140.252.13.35.6666

The first thing we notice in the interactive output shown earlier on bsdi is that only two datagrams
were received by the application: the first one from sun with all 1s, and the first one from svr4 with
all 4s. The other four datagrams appear to have been thrown away.

The tcpdump output in Figure 11.20 shows that all six datagrams were delivered to the destination
host. The datagrams were typed on the two clients in alternating order: first from sun, then from
svr4, and so on. We can also see that all six were delivered in about 12 seconds, within the 30-second
period while the server was sleeping.

1 0.0 sun.1252 > 140.252.13.63.6666: udp 11
2 2.499184 (2.4992) svr4.1042 > bsdi.6666: udp 14
3 4.959166 (2.4600) sun.1252 > 140.252.13.63.6666: udp 10
4 7.607149 (2.6480) svr4.1042 > bsdi.6666: udp 16
5 10.079059 (2.4719) sun.1252 > 140.252.13.63.6666: udp 12
6 12.415943 (2.3369) svr4.1042 > bsdi.6666: udp 9

Figure 11.20 tcpdump for UDP datagrams sent by two clients.

We can also see the server’s -E option lets it know the destination IP address of each datagram. If it
wanted to, it could choose what to do with the first datagram it receives, which was sent to a broadcast
address.

We can see several points in this example. First, the application is not told when its input queue
overflows. The excess datagrams are just discarded by UDP. Also, from the tcpdump output we see
that nothing is sent back to the client to tell it that its datagram was discarded. There is nothing like an
ICMP source quench sent back to the sender. Finally, it appears that the UDP input queue is FIFO
(first-in, first-out), whereas we saw that the ARP input queue in Section 11.9 was LIFO (last-in,
first-out).

161

Restricting Local IP Address
Most UDP servers wildcard their local IP address when they create a UDP end point. This means that
an incoming UDP datagram destined for the server’s port will be accepted on any local interface. For
example, we can start a UDP server on port 7777:

sun % sock -u -s 7777

We then use the netstat command to see the state of the end point:

sun % netstat -a -n -f inet
Active Internet connections (including servers)
Proto Recv-Q Send-Q Local Address Foreign Address (state)
udp 0 0 *.7777 *.*

We have deleted many lines of output other than the one in which we’re interested. The -a flag
reports on all network end points. The -n flag prints IP addresses as dotted-decimal numbers, instead
of trying to use the DNS to convert the address to a name, and prints numeric port numbers instead of
service names. The -f inet option reports only TCP and UDP end points.

The local address is printed as *.7777 where the asterisk means the local IP address has been
wildcarded.

When the server creates its end point it can specify one of the host’s local IP addresses, including one
of its broadcast addresses, as the local IP address for the end point. Incoming UDP datagrams will then
be passed to this end point only if the destination IP address matches the specified local address. With
our sock program, if we specify an IP address before the port number, that IP address becomes the
local IP address for the end point. For example,

sun % sock -u -s 140.252.1.29 7777

restricts the server to datagrams arriving on the SLIP interface (140.252.1.29). The netstat output
shows this:

Foreign
Proto Recv-Q Send-Q Local Address (state)
Address
udp 0 0 140.252.1.29.7777 *.*

If we try to send this server a datagram from a host on the Ethernet, bsdi at address 140.252.13.35,
an ICMP port unreachable is returned. The server never sees the datagram. Figure 11.21 shows this
scenario.

1 0.0 bsdi.1723 > sun.7777: udp 13
0.000822 sun > bsdi: icmp: sun udp port 7777
2

162

Figure 11.21 Rejection of UDP datagram caused by server’s local address binding.

It is possible to start different servers at the same port, each with a different local IP address.
Normally, however, the system must be told by the application that it is OK to reuse the same port
number.

With the sockets API the SO_REUSEADDR socket option must be specified. This is done by our sock
program by specifying the -A option.

On our host sun we can start five different servers on the same UDP port (8888):

sun % sock -u -s 140.252.1.29 8888 for SLIP link
sun % sock -u -s -A 140.252.13.33 8888 for Ethernet
for loopback
sun % sock -u -s -A 127.0.0.1 8888
interface
for Ethernet
sun % sock -u -s -A 140.252.13.63 8888
broadcasts
sun % sock -u -s -A 8888 everything else (wildcard
IP address)

All except the first of the servers must be started with the -A flag, telling the system that it’s OK to
reuse the same port number. The netstat output shows the five servers:

Foreign
Address
udp 0 0 *.8888 *.*
udp 0 0 140.252.13.63.8888 *.*
udp 0 0 127.0.0.1.8888 *.*
udp 0 0 140.252.13.33.8888 *.*
udp 0 0 140.252.1.29.8888 *.*

In this scenario, the only datagrams that will go to the server with the wildcarded local IP address are
those destined to 140.252.1.255, because the other four servers cover all other possibilities.

There is a priority implied when an end point with a wildcard address exists. An end point with a
specific IP address that matches the destination IP address is always chosen over a wildcard. The
wildcard end point is used only when a specific match is not found.

Restricting Foreign IP Address
In all the netstat output that we showed earlier, the foreign IP address and foreign port number are
shown as *.* meaning the end point will accept an incoming UDP datagram from any IP address and
any port number. Most implementations allow a UDP end point to restrict the foreign address.

163

This means the end point will only receive UDP datagrams from that specific IP address and port
number. Our sock program uses the -f option to specify the foreign IP address and port number:

sun % sock -u -s -f 140.252.13.35.4444 5555

This sets the foreign IP address to 140.252.13.35 (our host bsdi) and the foreign port number to
4444. The server’s well-known port is 5555. If we run netstat we see that the local IP address has
also been set, even though we didn’t specify it:

Foreign
Address
udp 0 0 140.252.13.33.5555 *.*

This is a side effect of specifying the foreign IP address and foreign port on Berkeley-derived systems:
if the local address has not been chosen when the foreign address is specified, the local address is
chosen automatically. Its value becomes the IP address of the interface chosen by IP routing to reach
the specified foreign IP address. Indeed, in this example the IP address on sun for the Ethernet that is
connected to the foreign address is 140.252.13.33.

Figure 11.22 summarizes the three types of address bindings that a UDP server can establish for itself.

Local Address Foreign Address Description
localIP.lport foreignIP.fport restricted to one client
localIP.lport *.* restricted to datagrams arriving on one local interface: localIP
*. lport *.* receives all datagrams sent to lport

Figure 11.22 Specification of local and foreign IP addresses and port number for UDP server.

In all cases, lport is the server’s well-known port and localIP must be the IP address of a local
interface. The ordering of the three rows in the table is the order that the UDP module applies when
trying to determine which local end point receives an incoming datagram. The most specific binding
(the first row) is tried first, and the least specific (the last row with both IP addresses wildcarded) is
tried last.

Multiple Recipients per Port
Although it’s not specified in the RFCs, most implementations allow only one application end point at
a time to be associated with any one local IP address and UDP port number. When a UDP datagram
arrives at a host destined for that IP address and port number, one copy is delivered to that single end
point. The IP address of the end point can be the wildcard, as shown earlier.

For example, under SunOS 4.1.3 we start one server on port 9999 with a wildcarded local IP address:

sun % sock -u -s 9999

If we then try to start another server with the same wildcarded local address and the same port, it
doesn’t work, even if we specify the -A option:

164

sun % sock -u -s 9999 we expect this to fail
can’t bind local address: Address already in use

sun % sock -u-s-A 9999 so we try -A flag this time

On systems that support multicasting (Chapter 12), this changes. Multiple end points can use the same
local IP address and UDP port number, although the application normally must tell the API that this is
OK (i.e., our -A flag to specify the SO_REUSEADDR socket option).

4.4BSD, which supports multicasting, requires the application to set a different socket option
(SO_REUSEPORT) to allow multiple end points to share the same port. Furthermore each end point
must specify this option, including the first one to use the port.

When a UDP datagram arrives whose destination IP address is a broadcast or multicast address, and
there are multiple end points at the destination IP address and port number, one copy of the incoming
datagram is passed to each end point. (The end point’s local IP address can be the wildcard, which
matches any destination IP address.) But if a UDP datagram arrives whose destination IP address is a
unicast address, only a single copy of the datagram is delivered to one of the end points. Which end
point gets the unicast datagram is implementation dependent.

11.13 Summary
UDP is a simple protocol. Its official specification, RFC 768 [Postel 1980], requires only three pages.
The services it provides to a user process, above and beyond IP, are port numbers and an optional
checksum. We used UDP to examine this checksum and to see how fragmentation is performed.

We then examined the ICMP unreachable error that is part of the new path MTU discovery feature
(Section 2.9). We watched path MTU discovery using Traceroute and UDP We also looked at the
interaction between UDP and ARP whereby most ARP implementations only retain the most recently
transmitted datagram to a given destination, while waiting for an ARP reply.

The ICMP source quench error can be sent by a system that is receiving IP datagrams faster than they
can be processed. It is easy to generate these ICMP errors using UDP.

Exercises
11.1 In Section 11.5 we caused fragmentation on an Ethernet by writing a UDP datagram with 1473
bytes of user data. What is the smallest amount of user data that causes fragmentation on an Ethernet if
IEEE 802 encapsulation (Section 2.2) is used instead?

11.2 Read RFC 791 [Postel 1981a] to determine why all fragments other than the last must have a
length that is a multiple of 8 bytes.

11.3 Assume an Ethernet and a UDP datagram with 8192 bytes of user data. How many fragments are
transmitted and what is the offset and length of each fragment?

11.4 Continue the previous exercise, assuming these fragments then traverse a SLIP link with an MTU
of 552. You also need to remember that the amount of data in each fragment (i.e., everything other
than the IP header) must be a multiple of 8 bytes. How many fragments are transmitted and what is the
offset and length of each fragment?

165

11.5 An application using UDP sends a datagram that gets fragmented into four pieces. Assume that
fragments 1 and 2 make it to the destination, with fragments 3 and 4 being lost. The application then
times out and retransmits the UDP datagram 10 seconds later and this datagram is fragmented
identically to the first transmission (i.e., same offsets and lengths). Assume that this time fragments 1
and 2 are lost but fragments 3 and 4 make it to the destination. Also assume that the reassembly timer
on the receiving host is 60 seconds, so when fragments 3 and 4 of the retransmission make it to the
destination, fragments 1 and 2 from the first transmission have not been discarded. Can the receiver
reassemble the IP datagram from the four fragments it now has?

11.6 How do you know that the fragments in Figure 11.15 really correspond to lines 5 and 6 in Figure
11.14?

11.7 After the host gemini had been up for 33 days, the netstat program showed that 129 IP
datagrams out of 48 mi1110n had been dropped because of a bad header checksum, and 20 TCP
segments out of 30 mi1110n had been dropped because of a bad TCP checksum. Not a single UDP
datagram was dropped, however, because of a UDP checksum error, out of the approximately 18
mi1110n UDP datagrams. Give two reasons why. (Hint: See Figure 11.4.)

11.8 In our discussion of fragmentation we never said what happens to IP options in the IP header-are
they copied as part of the IP header in each fragment, or left in the first fragment only? We’ve
described the following IP options: record route (Section 7.3), time-stamp (Section 7.4), strict and
loose source routing (Section 8.5). How would you expect fragmentation to handle these options?
Check your answer with RFC 791.

11.9 In Figure 1.8 we said that incoming UDP datagrams are demultiplexed based on the destination
UDP port number. Is that correct?

166

Broadcasting and Multicasting
12.1 Introduction
We mentioned in Chapter 1 that there are three kinds of IP addresses: unicast, broadcast, and
multicast. In this chapter we discuss broadcasting and multicasting in more detail.

Broadcasting and multicasting only apply to UDP, where it makes sense for an application to send a
single message to multiple recipients. TCP is a connection-oriented protocol that implies a connection
between two hosts (specified by IP addresses) and one process on each host (specified by port
numbers).

Consider a set of hosts on a shared network such as an Ethernet. Each Ethernet frame contains the
source and destination Ethernet addresses (48-bit values). Normally each Ethernet frame is destined
for a single host. The destination address specifies a single interface-called a unicast. In this way
communication between any two hosts doesn’t bother any of the remaining hosts on the cable (except
for possible contention for the shared media).

There are times, however, when a host wants to send a frame to every other host on the cable-called a
broadcast. We saw this with ARP and RARP. Multicasting fits between unicasting and broadcasting:
the frame should be delivered to a set of hosts that belong to a multicast group.

To understand broadcasting and multicasting we need to understand that filtering takes place on each
host, each time a frame passes by on the cable. Figure 12.1 shows a picture of this.

First, the interface card sees every frame that passes by on the cable and makes a decision whether to
receive the frame and deliver it to the device driver. Normally the interface card receives only those
frames whose destination address is either the hardware address of the interface or the broadcast
address. Additionally, most interfaces can be placed into a promiscuous mode whereby they receive a
copy of every frame. This mode is used by tcpdump, for example.

Figure 12.1 Filtering that takes place up the protocol stack when a frame is received.

Today most interfaces can also be configured to receive frames whose destination address is a
multicast address, or some subset of multicast addresses. On an Ethernet, a multicast address has the
low-order bit of the high-order byte turned on. In hexadecimal this bit looks like
01:00:00:00:00:00. (We can consider the Ethernet broadcast address, ff:ff:ff:ff:ff:ff
as a special case of the Ethernet multicast address.)

If the interface card receives the frame, it is passed to the device driver. (One reason the interface card
might discard the frame is if the Ethernet checksum is incorrect.) Additional filtering is performed by
the device driver. First, the frame type must specify a protocol that is supported (IP, ARP, etc.).
Second, additional multicast filtering may be performed, to check whether the host belongs to the
addressed multicast group.

167

The device driver then passes the frame to the next layer, such as IP, if the frame type specifies an IP
datagram. IP performs more filtering, based on the source and destination IP addresses, and passes the
datagram up to the next layer (such as TCP or UDP) if all is well.

Each time UDP receives a datagram from IP, it performs filtering based on the destination port
number, and sometimes the source port number too. If no process is currently using the destination
port number, the datagram is discarded and an ICMP port unreachable message is normally generated.
(TCP performs similar filtering based on its port numbers.) If the UDP datagram has a checksum error,
UDP silently discards it.

The problem with broadcasting is the processing load that it places on hosts that aren’t interested in the
broadcasts. Consider an application that is designed to use UDP broadcasts. If there are 50 hosts on the
cable, but only 20 are participating in the application, every time one of the 20 sends a UDP broadcast,
the other 30 hosts have to process the broadcast, all the way up through the UDP layer, before the
UDP datagram is discarded. The UDP datagram is discarded by these 30 hosts because the destination
port number is not in use.

The intent of multicasting is to reduce this load on hosts with no interest in the application. With
multicasting a host specifically joins one or more multicast groups. If possible, the interface card is
told which multicast groups the host belongs to, and only those multicast frames are received.

12.2 Broadcasting
In Figure 3.9 we showed four different forms of IP broadcast addresses. We now describe them in
more detail.

Limited Broadcast
The limited broadcast address is 255.255.255.255. This can be used as the destination address of an IP
datagram during the host configuration process, when the host might not know its subnet mask or even
its IP address.

A datagram destined for the limited broadcast address is never forwarded by a router under any
circumstance. It only appears on the local cable.

An unanswered question is: if a host is multihomed and a process sends a datagram to the limited
broadcast address, should the datagram be sent out each connected interface that supports
broadcasting? If not, an application that wants to broadcast out all interfaces must determine all the
interfaces on the host that support broadcasting, and send a copy out each interface.

Most BSD systems treat 255.255.255.255 as an alias for the broadcast address of the first interface that
was configured, and don’t provide any way to send a datagram out all attached, broadcast-capable
interfaces. Indeed, two applications that send UDP datagrams out every interface are routed
(Section 10.3) and rwhod (the server for the BSD rwho client). Both of these applications go through
a similar start-up procedure to determine all the interfaces on the host, and which ones are capable of
broadcasting. The net-directed broadcast address corresponding to that interface is then used as the
destination address for datagrams sent out the interface.

The Host Requirements RFC takes no stand on the issue of whether a multihomed host should send a limited broadcast out
all its interfaces.

168

Net-directed Broadcast
The net-directed broadcast address has a host ID of all one bits. A class A net-directed broadcast
address is netid.255.255.255, where netid is the class A network ID.

A router must forward a net-directed broadcast, but it must also have an option to disable this
forwarding.

Subnet-directed Broadcast
The subnet-directed broadcast address has a host ID of all one bits but a specific subnet ID.
Classification of an IP address as a subnet-directed broadcast address requires knowledge of the subnet
mask. For example, if a router receives a datagram destined for 128.1.2.255, this is a subnet-directed
broadcast if the class B network 128.1 has a subnet mask of 255.255.255.0, but it is not a broadcast if
the subnet mask is 255.255.254.0 (0xfffffe00).

All-subnets-directed Broadcast
An all-subnets-directed broadcast address also requires knowledge of the destination network’s
subnet mask, to differentiate this broadcast address from a net-directed broadcast address. Both the
subnet ID and the host ID are all one bits. For example, if the destination’s subnet mask is
255.255.255.0, then the IP address 128.1.255.255 is an all-subnets-directed broadcast. But if the
network is not subnetted, then this is a net-directed broadcast.

Current feeling [Almquist 1993] is that this type of broadcast is obsolete. It is better to use
multicasting than an all-subnets-directed broadcast.

[Almquist 1993] notes that RFC 922 requires that an all-subnets-directed broadcast be sent to all subnets, but no current
routers do so. This is fortunate since a host that has been misconfigured without its subnet mask sends all its "local"
broadcasts to all subnets. For example, if the host with IP address 128.1.2.3 doesn’t set a subnet mask, then its broadcast
address normally defaults to 128.1.255.255. But if the subnet mask should have been set to 255.255.255.0, then broadcasts
from this misconfigured host appear directed to all subnets.

The first widespread implementation of TCP/IP, the 4.2BSD system in 1983, used a host ID of all zero bits for the
broadcast address. One of the earliest references to the broadcast IP address is IEN 212 [Gurwitz and Hinden 1982], and it
proposed to define the IP broadcast address as a host ID of one bits. (lENs are the Internet Experiment Notes, basically
predecessors to the RFCs.) RFC 894 [Hornig 1984] commented that 4.2BSD used a nonstandard broadcast address, but
RFC 906 [Finlayson 1984] noted that there was no Internet standard for the broadcast address. The RFC editor added a
footnote to RFC 906 acknowledging the lack of a standard broadcast address, but strongly recommended that a host ID of
all one bits be used as the broadcast address. Although Berkeley adopted the use of all one bits for the broadcast address
with 4.3BSD in 1986, some operating systems (notably SunOS 4.x) continued to use the nonstandard broadcast address
through the early 1990s.

12.3 Broadcasting Examples
How are broadcasts sent and what do routers and hosts do with broadcasts? Unfortunately this is a
hard question to answer because it depends on the type of broadcast address, the application, the
TCP/IP implementation, and possible configuration switches. First, the application must support
broadcasting. If we execute

sun % ping 255.255.255.255
/usr/etc/ping: unknown host 255.255.255.255

169

intending to send a broadcast on the local cable, it doesn’t work. But the problem here is a
programming problem in the application (ping). Most applications that accept either a dotted-decimal
IP address or a hostname call the function inet_addr(3) to convert the dotted-decimal character
string to its 32-bit binary IP address, and if this fails, assume the character string is a hostname.
Unfortunately this library function returns -1 to indicate an error (such as a character other than a digit
or decimal point in the string), but the limited broadcast address (255.255.255.255) also converts into
-1. Most programs then assume that the character string is a hostname, look it up using the DNS
(Chapter 14), and end up printing an error such as "unknown host."

If we fix this programming shortfall in the ping program, however, the results are often not what we
expect. On six different systems tested by the author, only one handled this as expected and generated
a broadcast packet on the local cable. Most looked up the IP address 255.255.255.255 in the routing
table, applied the default route, and sent a unicast packet to the default router. Eventually the packet
was thrown away

A subnet-directed broadcast is what we should be using. Indeed, in Section 6.3 we sent datagrams to
the IP address 140.252.13.63 for the bottom Ethernet in our test network (inside front cover), and got
replies from all the hosts on the Ethernet. The subnet-directed broadcast address associated with each
interface is the value used with the ifconfig command (Section 3.8). If we ping that address, the
result is what we expect:

sun % arp -a ARP cache is empty

sun % ping 140.252.13.63
PING 140.252.13.63: 56 data bytes
64 bytes from sun (140.252.13.33): icmp_seq=0. time=4. ms
64 bytes frombsdi (140.252.13.35): icmp_seq=0. time=172. ms

64 bytes from sun (140.252.13.33): icmp_seq=l. time=1. ms
64 bytes from bsdi (140.252.13.35): icmp_seq=l. time=52. ms
64 bytes from svr4 (140.252.13.34): icmp_seq=l. time=90. ms

--140.252.13.63 PING Statistics--
2 packets transmitted, 6 packets received, -200% packet loss

sun % arp -a check ARP cache again
svr4 (140.252.13.34) at 0:0:c0:c2:9b:26
bsdi (140.252.13.35) at 0:0:c0:6f:2d:40

IP looks at the destination address (140.252.13.63), determines that it is the subnet-directed broadcast
address, and sends the datagram to the link-layer broadcast address.

We mentioned in Section 6.3 that this type of broadcast means all the hosts on the local network,
including the sender. We see here that we do get a reply from the sending host (sun) in addition to the
other hosts on the cable.

In this example we’ve also shown the ARP cache before and after the ping of the broadcast address.
This is to show the interaction between broadcasting and ARP. "The ARP cache is empty before we
execute ping, but full afterward. (That is, there is one entry for every other host on the cable that

170

responded to the echo request.) How did this happen when we said that the Ethernet frame is sent to
the link-layer broadcast address (0xffffffff)? The sending of these frames by sun does not require
ARP.

If we watch ping using tcpdump, we see that it is the recipients of the broadcast frames that
generate an ARP request to sun, before they can send their reply. This is because the reply is unicast.
We said in Section 4.5 that the receiver of an ARP request (sun in this example) normally adds the
requestor’s IP address and hardware address to its ARP cache, in addition to sending an ARP reply.
This is on the assumption that if the requestor is about to send us a packet, we’ll probably want to send
something back.

Our use of ping is somewhat special because the type of programming interface that it uses (called
"raw sockets" on most Unix implementations) always allows a datagram to be sent to the broadcast
address. What if we use an application that was not designed to support broadcasting, such as TFTP?
(We cover TFTP in more detail in Chapter 15.)

bsdi % tftp start the client
tftp> connect 140.252.13.63 specify the IP address of the server
tftp> get temp.foo and try to fetch a file from the server
tftp: sendto: Permission denied
tftp> quit terminate the client

Here we get an error immediately, and nothing is sent on the cable. What’s happening here is that the
sockets API doesn’t allow a process to send a UDP datagram to the broadcast address unless the
process specifically states that it plans to broadcast. This is intended to prevent users from mistakenly
specifying a broadcast address (as we did here) when the application was never intended to broadcast.

With the sockets API the application must set the SO_BROADCAST socket option before sending a UDP datagram to a
broadcast address.

Not all systems enforce this restriction. Some implementations allow any process to broadcast UDP datagrams, without
requiring the process to say so. Others are more restrictive and require a process to have superuser privileges to broadcast.

The next question is whether directed broadcasts are forwarded or not. Some kernels and routers have
an option to enable or disable this feature. (See Appendix E.)

If we enable this feature on our router bsdi and run ping from the host slip, we can see if the
subnet-directed broadcasts are forwarded by bsdi. Forwarding a directed broadcast means the router
takes the incoming unicast datagram, determines that the destination address is the directed broadcast
for one of its interfaces, and then forwards the datagram onto the appropriate network using a
link-layer broadcast.

slip % ping 140.252.13.63

PING 140.252.13.63 (140.252.13.63): 56 data bytes 64 bytes
from 140.252.13.35: icmp_seq=0 ttl=255 time=190 ms 64 bytes
from 140.252.13.33: icmp_seq=0 ttl=254 time=280 ms (DUP!)
64 bytes from 140.252.13.34; icmp_seq=0 ttl=254 time=360 ms (DUP!)

171

64 bytes from 140.252.13.35; icmp_seq=l ttl=255 time=180 ms
64 bytes from 140.252.13.33: icmp_seq=l ttl=254 time=270 ms (DUP!)
64 bytes from 140.252.13.34: icmp_seq=l ttl=254 time=360 ms (DUP!)

- 140.252.13.63 ping statistics -
3 packets transmitted, 2 packets received, +4 duplicates, 33% packet
loss

We see that this does indeed work. We also see that the BSD ping program checks for duplicate
sequence numbers and prints DUP! when this occurs. It normally means a packet was duplicated
somewhere, but here we expect to see this, since we sent the requests to a broadcast address.

We can also run this test from a host much farther away from the network to which the broadcast is
directed. If we run ping from the host vangogh.cs.berkeley.edu (14 hops away from our
network), it still works if the router sun is configured to forward directed broadcasts. In this case the
IP datagrams (carrying the ICMP echo requests) are forwarded by every router in the path as a normal
datagram. None of them knows that it’s really a directed broadcast. The next to last router, netb,
thinks it’s for the host with an ID of 63, and forwards it to sun. It is the router sun that detects that
the destination IP address is really the broadcast address of an attached interface, and turns the
datagram into a link-layer broadcast on that network.

Broadcasting is a feature that should be used with great care. In many cases IP multicasting will prove
to be a better solution.

12.4 Multicasting
IP multicasting provides two services for an application.

1. Delivery to multiple destinations. There are many applications that deliver information to
multiple recipients: interactive conferencing and dissemination of mail or news to multiple
recipients, for example. Without multicasting these types of services tend to use TCP today
(delivering a separate copy to each destination). Even with multicasting, some of these
applications might continue to use TCP for its reliability.
2. Solicitation of servers by clients. A diskless workstation, for example, needs to locate a bootstrap
server. Today this is provided using a broadcast (as we’ll see with BOOTP in Chapter 16), but a
multicast solution would impose less overhead on the hosts that don’t provide the service.

In this section we’ll take a look at multicast addresses, and the next chapter looks at the protocol used
by multicasting hosts and routers (IGMP).

Multicast Group Addresses
Figure 12.2 shows the format of a class D IP address.

172

Figure 12.2 Format of a class D IP address.

Unlike the other three classes of IP addresses (A, B, and C), which we showed in Figure 1.5, the 28
bits allocated for the multicast group ID have no further structure.

A multicast group address is the combination of the high-order 4 bits of 1110 and the multicast group
ID. These are normally written as dotted-decimal numbers and are in the range 224.0.0.0 through
239.255.255.255.

The set of hosts listening to a particular IP multicast address is called a host group. A host group can
span multiple networks. Membership in a host group is dynamic-hosts may join and leave host groups
at will. There is no restriction on the number of hosts in a group, and a host does not have to belong to
a group to send a message to that group.

Some multicast group addresses are assigned as well-known addresses by the IANA (Internet
Assigned Numbers Authority). "These are called permanent host groups. This is similar to the
well-known TCP and UDP port numbers. Similarly, these well-known multicast addresses are listed in
the latest Assigned Numbers RFC. Notice that it is the multicast address of the group that is
permanent, not the membership of the group.

For example, 224.0.0.1 means "all systems on this subnet," and 224.0.0.2 means "all routers on this
subnet." The multicast address 224.0.1.1 is for NTP, the Network Time Protocol, 224.0.0.9 is for
RIP-2 (Section 10.5), and 224.0.1.2 is for SGI’s (Silicon Graphics) dogfight application.

Converting Multicast Group Addresses to Ethernet Addresses
The IANA owns an Ethernet address block, which in hexadecimal is 00:00:5e. This is the
high-order 24 bits of the Ethernet address, meaning that this block includes addresses in the range
00:00:5e:00:00:00 through 00:00:5e:ff:ff:ff. The IANA allocates half of this block for
multicast addresses. Given that the first byte of any Ethernet address must be 01 to specify a multicast
address, this means the Ethernet addresses corresponding to IP multicasting are in the range
01:00:5e:00:00:00 through 01:00:5e:7f:ff:ff.

Our notation here uses the Internet standard bit order, for a CSMA/CD or token bus network, as the bits appear in memory.
This is what most programmers and system administrators deal with. The IEEE documentation uses the transmission order
of the bits. The Assigned Numbers RFC gives additional details on the differences between these representations.

This allocation allows for 23 bits in the Ethernet address to correspond to the IP multicast group ID.
The mapping places the low-order 23 bits of the multicast group ID into these 23 bits of the Ethernet
address. This is shown in Figure 12.3.

Since the upper 5 bits of the multicast group ID are ignored in this mapping, it is not unique.
Thirty-two different multicast group IDs map to each Ethernet address. For example, the multicast
addresses 224.128.64.32 (hex e0.80.40.20) and 224.0.64.32 (hex e0.00 40.20) both map into
the Ethernet address 01:00:5e:00:40:20.

173

Since the mapping is not unique, it implies that the device driver or the IP module in Figure 12.1 must
perform filtering, since the interface card may receive multicast frames in which the host is really not
interested. Also, if the interface card doesn’t

Figure 12.3 Mapping of a class D IP address into Ethernet multicast address.

provide adequate filtering of multicast frames, the device driver may have to receive all multicast
frames, and perform the filtering itself.

LAN interface cards tend to come in two varieties. One type performs multicast filtering based on the hash value of the
multicast hardware address, which means some unwanted frames can always get through. The other type has a small, fixed
number of multicast addresses to listen for, meaning that when the host needs to receive more multicast addresses than are
supported, the interface must be put into a "multicast promiscuous" mode. Hence, both types of interfaces still require that
the device driver perform checking that the received frame is really wanted.

Even if the interface performs perfect multicast filtering (based on the 48-bit hardware address), since the mapping from a
class D IP address to a 48-bit hardware address is not one-to-one, filtering is still required.

Despite this imperfect address mapping and hardware filtering, multicasting is still better than broadcasting.

Multicasting on a single physical network is simple. The sending process specifies a destination IP
address that is a multicast address, the device driver converts this to the corresponding Ethernet
address, and sends it. The receiving processes must notify their IP layers that they want to receive
datagrams destined for a given multicast address, and the device driver must somehow enable
reception of these multicast frames. This is called "joining a multicast group." (The reason we use the
plural "receiving processes" is because there are normally multiple receivers for a given multicast
message, either on the same host or on multiple hosts, which is why we’re using multicasting in the
first place.) When a multicast datagram is received by a host, it must deliver a copy to all the processes
that belong to that multicast group. This is different from UDP where a single process receives an
incoming unicast UDP datagram. With multicasting it is possible for multiple processes on a given
host to belong to the same multicast group.

But complications arise when we extend multicasting beyond a single physical network and pass
multicast packets through routers. A protocol is needed for multicast routers to know if any hosts on a
given physical network belong to a given multicast group. This protocol is called the Internet Group
Management Protocol (IGMP) and is the topic of the next chapter.

Multicasting on FDDI and Token Ring Networks
FDDI networks use the same mapping between the class D IP address and the 48-bit FDDI address
[Katz 1990]. Token ring networks normally use a different mapping, because of limitations in most
token ring controllers [Pusateri 1993].

174

12.5 Summary
Broadcasting is sending a packet to all hosts on a network (usually a locally attached network) and
multicasting is sending a packet to a set of hosts on a network. Basic to these two concepts is an
understanding of the different types of filtering that occur when a received frame passes up a protocol
stack. Each layer can discard a received packet for different reasons.

There are four types of broadcast addresses: limited, net-directed, subnet-directed, and
all-subnets-directed. The most common is subnet-directed. The limited broadcast address is normally
seen only when a system is bootstrapping.

Problems occur when trying to broadcast through routers, often because the router may not know the
subnet mask of the destination network. The results depend on many factors: which type of broadcast
address, configuration parameters, and so on.

A class D IP address is called a multicast group address. It is converted to an Ethernet address by
placing its lower 23 bits into a fixed Ethernet address. The mapping is not unique, requiring additional
filtering by one of the protocol modules.

Exercises
12.1 Does broadcasting increase the amount of network traffic?

12.2 Consider 50 hosts on an Ethernet: 20 running TCP/IP and 30 running some other protocol suite.
How are broadcasts from one protocol suite handled by hosts running the other protocol suite?

12.3 You login to a Unix system that you’ve never used before and want to find the subnet-directed
broadcast address for all attached interfaces that support broadcasting. How can you do this?

12.4 If we ping the broadcast address with a large packet size, as in

sun % ping 140.252.13.63 1472
PING 140.252.13.63; 1472 data bytes
1480 bytes from sun (140.252.13.33): icmp_seq=0. time=6. ms
1480 bytes from bsdi (140.252.13.35): icmp_seq=0. time=128. ms

it works, but increasing the packet size by 1 byte gives us the following error:

sun % ping 140.252.13.63 1473
PING 140.252.13.63: 1473 data bytes
sendto: Message too long

What’s going on?

12.5 Redo Exercise 10.6 assuming the eight RIP messages are multicast instead of broadcast (assume
RIP Version 2 is being used). What changes?

175

IGMP: Internet Group Management Protocol
13.1 Introduction
Section 12.4 provided an overview of IP multicasting and described how class D IP addresses are
mapped into Ethernet addresses. We briefly mentioned how multicasting occurs on a single physical
network, but said complications occur when multiple networks are involved and the multicast
datagrams must pass through routers.

In this chapter we’ll look at the Internet Group Management Protocol (IGMP), which is used by hosts
and routers that support multicasting. It lets all the systems on a physical network know which hosts
currently belong to which multicast groups. This information is required by the multicast routers, so
they know which multicast datagrams to forward onto which interfaces. IGMP is defined in RFC 1112
[Deering 1989].

Like ICMP, IGMP is considered part of the IP layer. Also like ICMP, IGMP messages are transmitted
in IP datagrams. Unlike other protocols that we’ve seen, IGMP has a fixed-size message, with no
optional data. Figure 13.1 shows the encapsulation of an IGMP message within an IP datagram.

Figure 13.1 Encapsulation of an IGMP message within an IP datagram.

IGMP messages are specified in the IP datagram with a protocol value of 2.

13.2 IGMP Message
Figure 13.2 shows the format of the 8-byte IGMP message.

Figure 13.2 Format of fields in IGMP message.

The IGMP version is 1. An IGMP type of 1 is a query sent by a multicast router, and 2 is a response
sent by a host. The checksum is calculated in the same manner as the ICMP checksum.

The group address is a class D IP address. In a query the group address is set to 0, and in a report it
contains the group address being reported. We’ll say more about it in the next section when we see
how IGMP operates.

176

13.3 IGMP Protocol
Joining a Multicast Group
Fundamental to multicasting is the concept of a process joining a multicast group on a given interface
on a host. (We use the term process to mean a program being executed by the operating system.)
Membership in a multicast group on a given interface is dynamic-it changes over time as processes
join and leave the group.

We imply here that a process must have a way of joining a multicast group on a given interface. A
process can also leave a multicast group that it previously joined. These are required parts of any API
on a host that supports multicasting. We use the qualifier "interface" because membership in a group is
associated with an interface. A process can join the same group on multiple interfaces.

The release of IP multicasting for Berkeley Unix from Stanford University details these changes for the sockets API. These
changes are also provided in Solaris 2.x and documented in the ip(7) manual pages.

Implied here is that a host identifies a group by the group address and the interface. A host must keep
a table of all the groups that at least one process belongs to, and a reference count of the number of
processes belonging to the group.

IGMP Reports and Queries
IGMP messages are used by multicast routers to keep track of group membership on each of the
router’s physically attached networks. The following rules apply.

1. A host sends an IGMP report when the first process joins a group. If multiple processes on a
given host join the same group, only one report is sent, the first time a process joins that group.
This report is sent out the same interface on which the process joined the group.
2. A host does not send a report when processes leave a group, even when the last process leaves a
group. The host knows that there are no members in a given group, so when it receives the next
query (next step), it won’t report the group.
3. A multicast router sends an IGMP query at regular intervals to see if any hosts still have
processes belonging to any groups. The router must send one query out each interface. The group
address in the query is 0 since the router expects one response from a host for every group that
contains one or more members on that host.
4. A host responds to an IGMP query by sending one IGMP report for each group that still contains
at least one process.

Using these queries and reports, a multicast router keeps a table of which of its interfaces have one or
more hosts in a multicast group. When the router receives a multicast datagram to forward, it forwards
the datagram (using the corresponding multicast link-layer address) only out the interfaces that still
have hosts with processes belonging to that group.

Figure 13.3 shows these two IGMP messages, reports sent by hosts, and queries sent by routers. The
router is asking each host to identify each group on that interface.

177

Figure 13.3 IGMP reports and queries.

We talk about the TTL field later in this section.

Implementation Details
There are many implementation details in this protocol that improve its efficiency. First, when a host
sends an initial IGMP report (when the first process joins a group), there’s no guarantee that the report
is delivered (since IP is used as the delivery service). Another report is sent at a later time. This later
time is chosen by the host to be a random value between 0 and 10 seconds.

Next, when a host receives a query from a router it doesn’t respond immediately, but schedules the
responses for later times. (We use the plural "responses" because the host must send one report for
each group that contains one or more members.) Since multiple hosts can be sending a report for the
same group, each schedules its response using random delays. Also realize that all the hosts on a
physical network receive all the reports from other hosts in the same group, because the destination
address of the report in Figure 13.3 is the group’s address. This means that, if a host is scheduled to
send a report, but receives a copy of the same report from another host, the response can be canceled.
This is because a multicast router doesn’t care how many hosts belong to the group-only whether at
least one host belongs to the group. Indeed, a multicast router doesn’t even care which host belongs to
a group. It only needs to know that at least one host belongs to a group on a given interface.

On a single physical network without any multicast routers, the only IGMP traffic is the reports issued
by the hosts that support IP multicasting, when the host joins a new group.

Time-to-Live Field
In Figure 13.3 we noted that the TTL field of the reports and queries is set to 1. This refers to the
normal TTL field in the IP header. A multicast datagram with an initial TTL of 0 is restricted to the
same host. By default, multicast datagrams are sent with a TTL of 1. This restricts the datagram to the
same subnet. Higher TTLs can be forwarded by multicast routers.

Recall from Section 6.2 that an ICMP error is never generated in response to a datagram destined to a
multicast address. Multicast routers do not generate ICMP "time exceeded" errors when the TTL
reaches 0.

Normally user processes aren’t concerned with the outgoing TTL. One exception, however, is the Traceroute program
(Chapter 8), which is based on setting the TTL field. Since multicasting applications must be able to set the outgoing TTL
field, this implies that the programming interface must provide this capability to user processes.

By increasing the TTL an application can perform an expanding ring search for a particular server.
The first multicast datagram is sent with a TTL of 1. If no response is received, a TTL of 2 is tried,
then 3, and so on. In this way the application locates the closest server, in terms of hops.

178

The special range of addresses 224.0.0.0 through 224.0.0.255 is intended for applications that never
need to multicast further than one hop. A multicast router should never forward a datagram with one of
these addresses as the destination, regardless of the TTL.

All-Hosts Group
In Figure 13.3 we also indicated that the router’s IGMP query is sent to the destination IP address of
224.0.0.1. This is called the all-hosts group address. It refers to all the multicast-capable hosts and
routers on a physical network. Each host automatically joins this multicast group on all
multicast-capable interfaces, when the interface is initialized. Membership in this group is never
reported.

13.4 An Example
Now that we’ve gone through some of the details of IP multicasting, let’s take a look at the messages
involved. We’ve added IP multicasting support to the host sun and will use some test programs
provided with the multicasting software to see what happens.

First we’ll use a modified version of the netstat command that reports multicast group membership
for each interface. (We showed the standard netstat -ni output for this host in Section 3.9.) In the
following output we show the lines corresponding to multicast groups in a bold font:

sun % netstat -nia
Name Mtu Network Address lpkts lerrs Opkts Oerrs Coll
le0 1500 140.252.13. 140.252.13.33 4370 0 4924 0 0
224.0.0.1
08;00:20:03:f6:42
01:00:5e:00:00:01
sl0 552 140.252.1 140.252.1.29 13587 0 15615 0 0
224.0.0.1
lo0 1536 127 127.0.0.1 1351 0 1351 0 0
224.0.0.1

The -n option prints IP addresses in numeric format (instead of trying to print them as names), -i
prints the interface statistics, and -a reports on all configured interfaces.

The second line of output for le0 (the Ethernet) shows that this interface belongs to the multicast
group 224.0.0.1 ("all hosts"), and two lines later the corresponding Ethernet address is shown:
01:00:5e:00:00:01. This is what we expect for the Ethernet address, given the mapping we
described in Section 12.4. We also see that the other two interfaces that support multicasting, the SLIP
link sl0 and the loopback interface lo0, also belong to the all-hosts group.

We must also show the IP routing table, as the normal routing table is used for multicast datagrams.
The bold entry shows that all datagrams for 224.0.0.0 are sent to the Ethernet:

179

sun % netstat -rn
Routing tables
140.252.13.65 140.252.13.35 UGH 0 32 le0
127.0.0.1 127.0.0.1 UH 1 381 lo0
140.252.1.183 140.252.1.29 UH 0 6 sl0
default 140.252.1.183 UG 0 328 sl0
224.0.0.0 140.252.13.33 U 0 66 le0
140.252.13.32 140.252.13.33 U 8 5581 le0

If you compare this routing table to the one shown in Section 9.2 for the router sun, you’ll see that the
multicast entry is the only change.

We now run a test program that lets us join a multicast group on an interface. (We don’t show any
output for our use of this test program.) We join the group 224.1.2.3 on the Ethernet interface
(140.252.13.33). Executing netstat shows that the kernel has joined the group, and again the
Ethernet address is what we expect. We show the changes from the previous netstat output in a
bold font:

sun % netstat -nia
Name Mtu Network Address lpkts lerrs Opkts Oerrs Coll
le0 1500 140.252.13. 140.252.13.33 4374 0 4929 0 0
224.1.2.3
224.0.0.1
08:00:20:03:f6:42
01:00:5e:01:02:03
01:00:5e:00:00:01
sl0 552 140.252.1 140.252.1.29 13862 0 15943 0 0
224.0.0.1
lo0 1536 127 127.0.0.1 1360 0 1360 0 0
224.0.0.1

We have shown the output again for the other two interfaces, sl0 and lo0, to reiterate that the
multicast group is joined only on one interface.

Figure 13.4 shows the tcpdump output corresponding to the process joining the multicast group.

180

1 0.0 8:0:20:3:f6:42 1:0:5e:1:2:3 ip 60:
sun > 224.1.2.3: igmp report 224.1.2.3 [ttl 1]
2 6.94 (6.94) 8:0:20:3:f6:42 1:0:5e:1:2:3 ip 60:
sun > 224.1.2.3: igmp report 224.1.2.3 [ttl 1]

Figure 13.4 tcpdump output when a host joins a multicast group.

Line 1 occurs when the host joins the group. The next line is the delayed report that we said is sent at
some random time up to 10 seconds afterward.

We have shown the hardware addresses in these two lines, to verify that the Ethernet destination
address is the correct multicast address. We can also see that the source IP address is the one
corresponding to sun, and the destination IP address is the multicast group address. We can also see
that the reported address is that same multicast group address.

Finally, we note that the TTL is 1, as specified, tcpdump prints the TTL in square brackets when its
value is 0 or 1. This is because the TTL is normally greater than this. With multicasting, however, we
expect to see lots of IP datagrams with a TTL of 1.

Implied in this output is that a multicast router must receive all multicast datagrams on all its
interfaces. The router has no idea which multicast groups the hosts might join.

Multicast Router Example
Let’s continue the previous example, but we’ll also start a multicast routing daemon on the host sun.
Our interest here is not the details of multicast routing protocols, but to see the IGMP queries and
reports that are exchanged. Even though the multicast routing daemon is running on the only host that
supports multicasting (sun), all the queries and reports are multicast on the Ethernet, so we can see
them on any other system on the Ethernet.

Before starting the routing daemon we joined another multicast group: 224.9.9.9. Figure 13.5 shows
the output.

181

1 0.0 sun > 224.0.0.4: igmp report 224.0.0.4
2 0.00 ( 0.00) sun > 224.0.0.1: igmp query
3 5.10 ( 5.10) sun > 224.9.9.9: igmp report 224.9.9.9
4 5.22 ( 0.12) sun > 224.0.0.1: igmp query
5 7.90 ( 2.68) sun > 224.1.2.3: igmp report 224.1.2.3
6 8.50 ( 0.60) sun > 224.0.0.4: igmp report 224.0.0.4
7 11.70 ( 3.20) sun > 224.9.9.9: igmp report 224.9.9.9
8 125.51 (113.81) sun > 224.0.0.1: igmp query
9 125.70 ( 0.19) sun > 224.9.9.9: igmp report 224.9.9.9
10 128.50 ( 2.80) sun > 224.1.2.3: igmp report 224.1.2.3
11 129.10 ( 0.60) sun > 224.0.0.4: igmp report 224.0.0.4
12 247.82 (118.72) sun > 224.0.0.1: igmp query
13 248.09 ( 0.27) sun > 224.1.2.3: igmp report 224.1.2.3
14 248.69 ( 0.60) sun > 224.0.0.4: igmp report 224.0.0.4
15 255.29 ( 6.60) sun > 224.9.9.9: igmp report 224.9.9.9

Figure 13.5 tcpdump output while multicast routing daemon is running.

We have not included the Ethernet addresses in this output, because we’ve already verified that they
are what we expect. We’ve also deleted the notation that the TTL equals 1, because again that’s what
we expect.

Line 1 is output when the routing daemon starts. It sends a report that it has joined the group 224.0.0.4.
Multicast address 224.0.0.4 is a well-known address used by DVMRP (Distance Vector Multicast
Routing Protocol), the protocol currently used for multicast routing. (DVMRP is defined in RFC 1075
[Waitzman, Partridge, and Deering 1988].)

When the daemon starts it also sends out a query (line 2). The destination IP address of the query is
224.0.0.1 (all-hosts), as shown in Figure 13.3.

The first report (line 3) is received about 5 seconds later, for group 224.9.9.9. This is the only report
received before another query is sent (line 4). These two queries (lines 2 and 4) occur rapidly when the
daemon starts up, as it tries to build its multicast routing table.

Lines 5, 6, and 7 are what we expect: one report from the host sun for each group to which it belongs.
Notice that the group 224.0.0.4 is reported, in addition to the two groups that we explicitly joined,
because as long as the routing daemon is running, it belongs to this group.

The next query on line 8 occurs about 2 minutes after the previous query. Again it elicits the three
reports we expect (lines 9, 10, and 11). The reports are in a different order this time, as expected, since
the time between receiving the query and sending the report should be randomized.

182

The final query that we show occurs about 2 minutes after the previous query, and again we have the
expected responses.

13.5 Summary
Multicasting is a way to send a message to multiple recipients. In many applications it is better than
broadcasting, since multicasting imposes less overhead on hosts that are not participating in the
communication. The simple host membership reporting protocol (IGMP) is the basic building block
for multicasting.

Multicasting on a LAN or across closely connected LANs uses the techniques we’ve described in this
chapter. Since broadcasting is often restricted to a single LAN, multicasting could be used instead of
broadcasting for many applications that use broadcasting today

A problem that has not been completely solved, however, is multicasting across wide area networks.
[Deering and Cheriton 1990] propose extensions to common routing protocols to support multicasting.
Section 9.13 of [Periman 1992] discusses some of the problems with multicasting across WANs.

[Casner and Deering 1992] describe the delivery of audio for an IETF meeting across the Internet
using multicasting and a virtual network called the MBONE (multicasting backbone).

Exercises
13.1 We said that hosts schedule IGMP reports with random delays. How can the hosts on a LAN try
to ensure that no two hosts generate the same random delay?

13.2 In [Casner and Deering 1992] they mention that UDP lacks two features needed for sending audio
samples across the MBONE: detection of packet reordering and detection of duplicate packets. How
could you add these capabilities above UDP?

183

DNS: The Domain Name System
14.1 Introduction
The Domain Name System, or DNS, is a distributed database that is used by TCP/IP applications to
map between hostnames and IP addresses, and to provide electronic mail routing information. We use
the term distributed because no single site on the Internet knows all the information. Each site
(university department, campus, company, or department within a company, for example) maintains its
own database of information and runs a server program that other systems across the Internet (clients)
can query. The DNS provides the protocol that allows clients and servers to communicate with each
other.

From an application’s point of view, access to the DNS is through a resolver. On Unix hosts the
resolver is accessed primarily through two library functions, gethostbyname(3) and
gethostbyaddr(3), which are linked with the application when the application is built. The first
takes a hostname and returns an IP address, and the second takes an IP address and looks up a
hostname. The resolver contacts one or more name servers to do the mapping.

In Figure 4.2 we showed that the resolver is normally part of the application. It is not part of the
operating system kernel as are the TCP/IP protocols. Another fundamental point from this figure is
that an application must convert a hostname to an IP address before it can ask TCP to open a
connection or send a datagram using UDP. The TCP/IP protocols within the kernel know nothing
about the DNS.

In this chapter we’ll take a look at how resolvers communicate with name servers using the TCP/IP
protocols (mainly UDP). We do not cover all the administrative details of running a name server or all
the options available with resolvers and servers. These details can fill an entire book. (See [Albitz and
Liu 1992] for all the details on the care and feeding of the standard Unix resolver and name server.)

RFC 1034 [Mockapetris 1987a] specifies the concepts and facilities provided by the DNS, and RFC
1035 [Mockapetris 1987b] details the implementation and specification. The most commonly used
implementation of the DNS, both resolver and name server, is called BIND-the Berkeley Internet
Name Domain server. The server is called named. An analysis of the wide-area network traffic
generated by the DNS is given in [Danzig, Obraczka, and Kumar 1992].

14.2 DNS Basics
The DNS name space is hierarchical, similar to the Unix filesystem. Figure 14.1 shows this
hierarchical organization.

184

Figure 14.1 Hierarchical organization of the DNS.

Every node (circles in Figure 14.1) has a label of up to 63 characters. The root of the tree is a special
node with a null label. Any comparison of labels considers uppercase and lowercase characters the
same. The domain name of any node in the tree is the list of labels, starting at that node, working up to
the root, using a period ("dot") to separate the labels. (Note that this is different from the Unix
filesystem, which forms a pathname by starting at the top and going down the tree.) Every node in the
tree must have a unique domain name, but the same label can be used at different points in the tree.

A domain name that ends with a period is called an absolute domain name or a fully qualified domain
name (FQDN). An example is sun.tuc.noao.edu.. If the domain name does not end with a
period, it is assumed that the name needs to be completed. How the name is completed depends on the
DNS software being used. If the uncompleted name consists of two or more labels, it might be
considered to be complete; otherwise a local addition might be added to the right of the name. For
example, the name sun might be completed by adding the local suffix .tuc.noao.edu.. The
top-level domains are divided into three areas:

1. arpa is a special domain used for address-to-name mappings. (We describe this in Section 14.5.)
2. The seven 3-character domains are called the generic domains. Some texts call these the
organizational domains.
3. All the 2-character domains are based on the country codes found in ISO 3166. These are called
the country domains, or the geographical domains.

185

Figure 14.2 lists the normal classification of the seven generic domains.

Domain Description
com commercial organizations
edu educational institutions
gov other U.S. governmental organizations
int international organizations
mil U.S. military
net networks
org other organizations

Figure 14.2 The 3-character generic domains.

DNS folklore says that the 3-character generic domains are only for U.S. organizations, and the
2-character country domains for everyone else, but this is false. There are many non-U.S.
organizations in the generic domains, and many U.S. organizations in the .us country domain. (RFC
1480 [Cooper and Postel 1993] describes the .us domain in more detail.) The only generic domains
that are restricted to the United States are .gov and .mil.

Many countries form second-level domains beneath their 2-character country code similar to the
generic domains: .ac.uk, for example, is for academic institutions in the United Kingdom and
.co.uk is for commercial organizations in the United Kingdom.

One important feature of the DNS that isn’t shown in figures such as Figure 14.1 is the delegation of
responsibility within the DNS. No single entity manages every label in the tree. Instead, one entity (the
NIC) maintains a portion of the tree (the top-level domains) and delegates responsibility to others for
specific zones.

A zone is a subtree of the DNS tree that is administered separately. A common zone is a second-level
domain, noao.edu, for example. Many second-level domains then divide their zone into smaller
zones. For example, a university might divide itself into zones based on departments, and a company
might divide itself into zones based on branch offices or internal divisions.

If you are familiar with the Unix filesystem, notice that the division of the DNS tree into zones is
similar to the division of a logical Unix filesystem into physical disk partitions. Just as we can’t tell
from Figure 14.1 where the zones of authority lie, we can’t tell from a similar picture of a Unix
filesystem which directories are on which disk partitions.

Once the authority for a zone is delegated, it is up to the person responsible for the zone to provide
multiple name servers for that zone. Whenever a new system is installed in a zone, the DNS
administrator for the zone allocates a name and an IP address for the new system and enters these into
the name server’s database. This is where the need for delegation becomes obvious. At a small
university, for example, one person could do this each time a new system was added, but in a large
university the responsibility would have to be delegated (probably by departments), since one person
couldn’t keep up with the work.

186

A name server is said to have authority for one zone or multiple zones. The person responsible for a
zone must provide a primary name server for that zone and one or more secondary name servers. The
primary and secondaries must be independent and redundant servers so that availability of name
service for the zone isn’t affected by a single point of failure.

The main difference between a primary and secondary is that the primary loads all the information for
the zone from disk files, while the secondaries obtain all the information from the primary. When a
secondary obtains the information from its primary we call this a zone transfer.

When a new host is added to a zone, the administrator adds the appropriate information (name and IP
address minimally) to a disk file on the system running the primary. The primary name server is then
notified to reread its configuration files. The secondaries query the primary on a regular basis
(normally every 3 hours) and if the primary contains newer data, the secondary obtains the new data
using a zone transfer.

What does a name server do when it doesn’t contain the information requested? It must contact
another name server. (This is the distributed nature of the DNS.) Not every name server, however,
knows how to contact every other name server. Instead every name server must know how to contact
the root name servers. As of April 1993 there were eight root servers and all the primary servers must
know the IP address of each root server. (These IP addresses are contained in the primary’s
configuration files. The primary servers must know the IP addresses of the root servers, not their DNS
names.) The root servers then know the name and location (i.e., the IP address) of each authoritative
name server for all the second-level domains. This implies an iterative process: the requesting name
server must contact a root server. The root server tells the requesting server to contact another server,
and so on. We’ll look into this procedure with some examples later in this chapter.

You can fetch the current list of root servers using anonymous FTP. Obtain the file
netinfo/root-servers.txt from either ftp.rs.internic.net or nic.ddn.mil.

A fundamental property of the DNS is caching. That is, when a name server receives information
about a mapping (say, the IP address of a hostname) it caches that information so that a later query for
the same mapping can use the cached result and not result in additional queries to other servers.
Section 14.7 shows an example of caching.

14.3 DNS Message Format
There is one DNS message defined for both queries and responses. Figure 14.3 shows the overall
format of the message.

187

Figure 14.3 General format of DNS queries and responses.

The message has a fixed 12-byte header followed by four variable-length fields.

The identification is set by the client and returned by the server. It lets the client match responses to
requests.

The 16-bit flags field is divided into numerous pieces, as shown in Figure 14.4.

Figure 14.4 flags field in the DNS header.

We’ll start at the leftmost bit and describe each field.

QR is a 1-bit field: 0 means the message is a query, 1 means it’s a response.
opcode is a 4-bit field. The normal value is 0 (a standard query). Other values are 1 (an inverse
query) and 2 (server status request).
AA is a 1-bit flag that means "authoritative answer." The name server is authoritative for the
domain in the question section.
TC is a 1-bit field that means "truncated." With UDP this means the total size of the reply
exceeded 512 bytes, and only the first 512 bytes of the reply was returned.
RD is a 1-bit field that means "recursion desired." This bit can be set in a query and is then
returned in the response. This flag tells the name server to handle the query itself, called a

188

recursive query. If the bit is not set, and the requested name server doesn’t have an authoritative
answer, the requested name server returns a list of other name servers to contact for the answer. This is
called an iterative query. We’ll see examples of both types of queries in later examples.
RA is a 1-bit field that means "recursion available." This bit is set to 1 in the response if the server
supports recursion. We’ll see in our examples that most name servers provide recursion, except for
some root servers.
There is a 3-bit field that must be 0.
rcode is a 4-bit field with the return code. The common values are 0 (no error) and 3 (name
error). A name error is returned only from an authoritative name server and means the domain name
specified in the query does not exist.

The next four 16-bit fields specify the number of entries in the four variable-length fields that
complete the record. For a query, the number of questions is normally 1 and the other three counts are
0. Similarly, for a reply the number of answers is at least 1, and the remaining two counts can be 0 or
nonzero.

Question Portion of DNS Query Message
The format of each question in the question section is shown in Figure 14.5. There is normally just one
question.

The query name is the name being looked up. It is a sequence of one or more labels. Each label begins
with a 1-byte count that specifies the number of bytes that follow. The name is terminated with a byte
of 0, which is a label with a length of 0, which is the label of the root. Each count byte must be in the
range of 0 to 63, since labels are limited.

Figure 14.5 Format of question portion of DNS query message.

to 63 bytes. (We’ll see later in this section that a count byte with the two high-order bits turned on,
values 192 to 255, is used with a compression scheme.) Unlike many other message formats that
we’ve encountered, this field is allowed to end on a boundary other than a 32-bit boundary. No
padding is used. Figure 14.6 shows how the domain name gemini.tuc.noao.edu is stored.

Figure 14.6 Representation of the domain name gemini.tuc.noao.edu.

189

Each question has a query type and each response (called a resource record, which we talk about
below) has a type. There are about 20 different values, some of which are now obsolete. Figure 14.7
shows some of these values. The query type is a superset of the type: two of the values we show can be
used only in questions.

Numeric query
Name Description type?
value type?
A 1 IP address * *
NS 2 name server * *
CNAME 5 canonical name * *
PTR 12 pointer record * *
HINFO 13 host info * *
MX 15 mail exchange record * *
AXFR 252 request for zone transfer *
* or ANY 255 request for all records *

Figure 14.7 type and query type values for DNS questions and responses.

The most common query type is an A type, which means an IP address is desired for the query name.
A PTR query requests the names corresponding to an IP address. This is a pointer query that we
describe in Section 14.5. We describe the other query types in Section 14.6.

The query class is normally 1, meaning Internet address. (Some other non-IP values are also supported
at some locations.)

Resource Record Portion of DNS Response Message
The final three fields in the DNS message, the answers,authority, and additional information fields,
share a common format called a resource record or RR. Figure 14.8 shows the format of a resource
record.

190

Figure 14.8 Format of DNS resource record.

The domain name is the name to which the following resource data corresponds. It is in the same
format as we described earlier for the query name field (Figure 14.6).

The type specifies one of the RR type codes. These are the same as the query type values that we
described earlier. The class is normally 1 for Internet data.

The time-to-live field is the number of seconds that the RR can be cached by the client. RRs often have
a TTL of 2 days.

The resource data length specifies the amount of resource data. The format of this data depends on
the type. For a type of 1 (an A record) the resource data is a 4-byte IP address.

Now that we’ve described the basic format of the DNS queries and responses, we’ll see what is passed
in the packets by watching some exchanges using tcpdump.

Let’s start with a simple example to see the communication between a resolver and a name server.
We’ll run the Telnet client on the host sun to the host gemini, connecting to the daytime server:

sun % telnet gemini daytime
Trying 140.252.1.11 ...
Connected to gemini. first three lines of output are from Telnet
tuc.noao.edu. client
Escape character is ’^]’ this is the output from the daytime server
Wed Mar 24 10:44:17 1993 and this is from the Telnet client
Connection closed by foreign
host.

191

For this example we direct the resolver on the host sun (where the Telnet client is run) to use the name
server on the host noao.edu (140.252.1.54). Figure 14.9 shows the arrangement of the three
systems.

Figure 14.9 Systems being used for simple DNS example.

As we’ve mentioned before, the resolver is part of the client, and the resolver contacts a name server to
obtain the IP address before the TCP connection can be established between Telnet and the daytime
server.

In this figure we’ve omitted the detail that the connection between sun and the 140.252.1 Ethernet is
really a SLIP link (see the figure on the inside front cover) because that doesn’t affect the discussion.
We will, however, run tcpdump on the SLIP link to see the packets exchanged between the resolver
and name server.

The file /etc/resolv.conf on the host sun tells the resolver what to do:

sun % cat /etc/resolv.conf
nameserver 140.252.1.54
domain tuc.noao.edu

The first line gives the IP address of the name server - the host noao.edu. Up to three
nameserver lines can be specified, to provide backup in case one is down or unreachable. The
domain line specifies the default domain. If the name being looked up is not a fully qualified domain
name (it doesn’t end with a period) then the default domain .tuc.noao.edu is appended to the
name. This is why we can type telnet gemini instead of telnet gemini.tuc.noao.edu.
Figure 14.10 shows the packet exchange between the resolver and name server.

140.252.1.29.1447 > 140.252.1.54.53: 1+ A?
1 0.0
gemini.tuc.noao.edu. (37)
140.252.1.54.53 > 140.252.1.29.1447: 1*
0.290820
2 2/0/0 A
(0.2908)
140.252.1.11 (69)

192

Figure 14.10 tcpdump output for name server query of the hostname gemini.tuc.noao.edu.

We’ve instructed tcpdump not to print domain names for the source and destination IP addresses of
each IP datagram. Instead it prints 140.252.1.29 for the client (the resolver) and 140.252.1.54 for the
name server. Port 1447 is the ephemeral port used by the client and 53 is the well-known port for the
name server. If tcpdump had tried to print names instead of IP addresses, then it would have been
contacting the same name server (doing pointer queries), confusing the output.

Starting with line 1, the field after the colon (1+) means the identification field is 1, and the plus sign
means the RD flag (recursion desired) is set. We see that by default, the resolver asks for recursion.

The next field, A?, means the query type is A (we want an IP address), and the question mark indicates
it’s a query (not a response). The query name is printed next: gemini.tuc.noao.edu.. The
resolver added the final period to the query name, indicating that it’s an absolute domain name.

The length of user data in the UDP datagram is shown as 37 bytes: 12 bytes are the fixed-size header
(Figure 14.3); 21 bytes for the query name (Figure 14.6), and 4 bytes for the query type and query
class. The odd-length UDP datagram reiterates that there is no padding in the DNS messages.

Line 2 in the tcpdump output is the response from the name server and 1* is the identification field
with the asterisk meaning the AA flag (authoritative answer) is set. (We expect this server, the primary
server for the noao.edu domain, to be authoritative for names within its domain.)

The output 2/0/0 shows the number of resource records in the final three variable-length fields in the
response: 2 answer RRs, 0 authority RRs, and 0 additional RRs. tcpdump only prints the first answer,
which in this case has a type of A (IP address) with a value of 140.252.1.11.

Why do we get two answers to our query? Because the host gemini is multihomed. Two IP addresses
are returned. Indeed, another useful tool with the DNS is a publicly available program named host. It
lets us issue queries to a name server and see what comes back. If we run this program we’ll see the
two IP addresses for this host:

sun % host gemini
gemini.tuc.noao.edu A 140.252.1.11
gemini.tuc.noao.edu A 140.252.3.54

The first answer in Figure 14.10 and the first line of output from the host command are the IP
address that shares the same subnet (140.252.1) as the requesting host. This is not an accident. If the
name server and the host issuing the query are on the same network (or subnet), then BIND sorts the
results so that addresses on common networks appear first.

We can still access the host gemini using the other address, but it might be less efficient. Using
traceroute in this instance shows that the normal route from subnet 140.252.1 to 140.252.3 is not
through the host gemini, but through another router that’s connected to both networks. So in this
case if we accessed gemini through the other IP address (140.252.3.54) all the packets would require
an additional hop. We return to this example and explore the reason for the alternative route in Section
25.9, when we can use SNMP to look at a router’s routing table.

193

There are other programs that provide easy interactive access to the DNS. nslookup is supplied with
most implementations of the DNS. Chapter 10 of [Albitz and Liu 1992] provides a detailed description
of how to use this program. The dig program ("Domain Internet Groper") is another publicly
available tool that queries DNS servers, doc ("Domain Obscenity Control") is a shell script that uses
dig and diagnoses misbehaving domains by sending queries to the appropriate DNS name servers,
and performing simple analysis of the responses. See Appendix F for details on how to obtain these
programs.

The final detail to account for in this example is the size of the UDP data in the reply: 69 bytes. We
need to know two points to account for these bytes.

1. The question is returned in the reply.
2. There can be many repetitions of domain names in a reply, so a compression scheme is used.
Indeed, in our example, there are three occurrences of the domain name
gemini.tuc.noao.edu.

The compression scheme is simple. Anywhere the label portion of a domain name can occur, the
single count byte (which is between 0 and 63) has its two high-order bits turned on instead. This
means it is a 16-bit pointer and not an 8-bit count byte. The 14 bits that follow in the pointer
specify an offset in the DNS message of a label to continue with. (The offset of the first byte in
the identification field is 0.) We purposely said that this pointer can occur wherever a label can
occur, not just where a complete domain name can. occur, since it’s possible for a pointer to form
either a complete domain name or just the ending portion of a name. (This is because the ending
labels in the names from a given domain tend to be identical.)

Figure 14.11 shows the format of the DNS reply, line 2 from Figure 14.10. We also show the IP and
UDP headers to reiterate that DNS messages are normally encapsulated in UDP datagrams. We
explicitly show the count bytes in the labels of the domain name in the question. The two answers
returned are the same, except for the different IP addresses returned in each answer. In this example
the pointer in each answer would have a value of 12, the offset from the start of the DNS header of the
complete domain name.

The final point to note from this example is from the second line of output from the Telnet command,
which we repeat here:

sun % telnet gemini daytime we only type gemini
Trying 140.252.1.11
... Connected to gemini.tuc.noao.edu. but the Telnet client outputs FQDN

We typed just the hostname (gemini), not the FQDN, but the Telnet client output the FQDN. What’s
happening is that the Telnet client looks up the name we type by calling

194

Figure 14.11 Format of DNS reply corresponding to line 2 of Figure 14.10.

the resolver (gethostbyname), which returns the IP addresses and the FQDN. Telnet then prints the
IP address that it’s trying to establish a TCP connection with, and when the connection is established,
it outputs the FQDN.

If there is a significant pause between typing the Telnet command and printing the IP address, this
delay is caused by the resolver contacting a name server to resolve the name into an IP address. A
pause between printing Trying and Connected to, however, is a delay caused by the
establishment of the TCP connection between the client and server, not the DNS.

14.5 Pointer Queries
A perpetual stumbling block in understanding the DNS is how pointer queries are handled - given an
IP address, return the name (or names) corresponding to that address.

First return to Figure 14.1 and examine the arpa top-level domain, and the in-addr domain
beneath it. When an organization joins the Internet and obtains authority for a portion of the DNS
name space, such as noao.edu, they also obtain authority for a portion of the in-addr.arpa
name space corresponding to their IP address on the Internet. In the case of noao.edu it is the class
B network ID 140.252. The level of the DNS tree beneath in-addr.arpa must be the first byte of
the IP address (140 in this example), the next level is the next byte of the IP address (252), and so on.
But remember that names are written starting at the bottom of the DNS tree, working upward. This
means the DNS name for the host sun, with an IP address of 140.252.13.33, is 33.13.252.140.
in-addr.arpa.

We have to write the 4 bytes of the IP address backward because authority is delegated based on
network IDs: the first byte of a class A address, the first and second bytes of a class B address, and the
first, second, and third bytes of a class C address. The first byte of the IP address must be immediately
below the in-addr label, but FQDNs are written from the bottom of the tree up. If FQDNs were
written from the top down, then the DNS name for the IP address would be
arpa.in-addr.140.252.13.33, but the FQDN for the host would be edu.noao.tuc.sun.

195

If there was not a separate branch of the DNS tree for handling this address-to-name translation, there
would be no way to do the reverse translation other than starting at the root of the tree and trying every
top-level domain. This could literally take days or weeks, given the current size of the Internet. The
in-addr.arpa solution is a clever one, although the reversed bytes of the IP address and the
special domain are confusing.

Having to worry about the in-addr.arpa domain and reversing the bytes of the IP address affects
us only if we’re dealing directly with the DNS, using a program such as host, or watching the
packets with tcpdump. From an application’s point of view, the normal resolver function
(gethostbyaddr) takes an IP address and returns information about the host. The reversal of the
bytes and appending the domain in-addr.arpa are done automatically by this resolver function.

Example
Let’s use the host program to do a pointer lookup and watch the packets with tcpdump. We’ll use
the same setup as in Figure 14.9, running the host program on the host sun, and the name server on
the host noao.edu. We specify the IP address of our host svr4:

sun % host 140.252.13.34
Name: svr4.tuc.noao.edu
Address: 140.252.13.34

Since the only command-line argument is an IP address, the host program automatically generates
the pointer query. Figure 14.12 shows the tcpdump output.

140.252.1.29.1610 > 140.252.1.54.53: 1+ PTR?
1 0.0
34.13.252.140.in-addr.arpa. (44)
140.252.1.54.53 > 140.252.1.29.1610: 1* 1/0/0
0.332288
2 PTR
(0.3323)
svr4.tuc.noao.edu. (75)

Figure 14.12 tcpdump output for a pointer query.

Line 1 shows that the identifier is 1, the recursion-desired flag is set (the plus sign), and the query type
is PTR. (Recall that the question mark means this is a query and not a response.) The data size of 44
bytes is from the 12-byte DNS header, 28 bytes for the 7 labels in the domain name, and 4 bytes for
the query type and query class.

The reply has the authoritative-answer bit set (the asterisk) and contains one answer RR. The RR type
is PTR and the resource data contains the domain name.

What is passed from the resolver to the name server for a pointer query is not a 32-bit IP address, but
the domain name 34.13.252.140.in-addr.arpa.

Hostname Spoofing Check
When an IP datagram arrives at a host for a server, be it a UDP datagram or a TCP connection request
segment, all that’s available to the server process is the client’s IP address and port number (UDP or
TCP). Some servers require the client’s IP address to have a pointer record in the DNS. We’ll see an
example of this, using anonymous FTP from an unknown IP address, in Section 27.3.

196

Other servers, such as the Rlogin server (Chapter 26), not only require that the client’s IP address have
a pointer record, but then ask the DNS for the IP addresses corresponding to the name returned in the
PTR response, and require that one of the returned addresses match the source IP address in the
received datagram. This check is because entries in the .rhosts file (Section 26.2) contain the
hostname, not an IP address, so the server wants to verify that the hostname really corresponds to the
incoming IP address.

Some vendors automatically put this check into their resolver routines, specifically the function
gethostbyaddr. This makes the check available to any program using the resolver, instead of
manually placing the check in each application.

We can see an example of this using the SunOS 4.1.3 resolver library. We have written a simple
program that performs a pointer query by calling the function gethostbyaddr. We have also set
our /etc/resolv.conf file to use the name server on the host noao.edu, which is across the
SLIP link from the host sun. Figure 14.13 shows the tcpdump output collected on the SLIP link
when the function gethostbyaddr is called to fetch the name corresponding to the IP address
140.252.1.29 (our host sun).

sun. 1812 > noao.edu.domain: 1+ PTR?
1 0.0
29.1.252.140.in-addr.arpa. (43)
noao.edu.domain > sun.1812: 1* 1/0/0 PTR
2 0.339091 (0.3391)
sun.tuc.noao.edu. (73)
sun. 1813 > noao.edu.domain: 2+ A?
3 0.344348 (0.0053)
sun.tuc.noao.edu. (33)
noao.edu.domain > sun.1813: 2* 2/0/0 A
4 0.669022 (0.3247)
140.252.1.29 (69)

Figure 14.13 Calling resolver function to perform pointer query

Line 1 is the expected pointer query, and line 2 is the expected response. But the resolver function
automatically sends an IP address query in line 3 for the name returned in line 2. The response in line
4 contains two answer records, since the host sun has two IP addresses. If one of the addresses does
not match the argument to gethostbyaddr, a message is sent to the system logging facility, and the
function returns an error to the application.

14.6 Resource Records
We’ve seen a few different types of resource records (RRs) so far: an IP address has a type of A, and
PTR means a pointer query. We’ve also seen that RRs are what a name server returns: answer RRs,
authority RRs, and additional information RRs. There are about 20 different types of resource records,
some of which we’ll now describe. Also, more RR types are being added over time.

A An A record defines an IP address. It is stored as a 32-bit binary value.
This is the pointer record used for pointer queries. The IP address is represented as a
PTR
domain name (a sequence of labels) in the in-addr.arpa domain.

197

This stands for "canonical name." It is represented as a domain name (a sequence of
labels). The domain name that has a canonical name is often called an alias. These are
used by some FTP sites to provide an easy to remember alias for some other system.

For example, the gated server (mentioned in Section 10.3) is available through
anonymous FTP from the server gated.cornell.edu. But there is no system named
CNAME gated, this is an alias for some other system. That other system is the canonical name
for gated.cornell.edu:

sun % host -t cname gated.cornell.edu
gated.cornell.edu CNAME COMET.CIT.CORNELL.EDO

Here we use the -t option to specify one particular query type.
Host information: two arbitrary character strings specifying the CPU and operating
system. Not all sites provide HINFO records for all their systems, and the information
provided may not be up to date.
HINFO
sun % host -t hinfo sun
sun.tuc.noao.edu HINFO Sun-4/25 Sun4.1.3

198

Mail exchange records, which are used in the following scenarios: (1) A site that is not
connected to the Internet can get an Internet-connected site to be its mail exchanger. The
two sites then work out an alternati ve way to exchange any mail that arrives, often using
the UUCP protocol. (2) MX records provide a way to deliver mail to an alternative host
when the destination host is not available. (3) MX records allow organizations to provide
virtual hosts that one can send mail to, such as cs.university.edu, even if a host
with that name doesn’t exist. (4) Organizations with firewall gateways can use MX
records to limit connectivity to internal systems.

Many sites that are not connected to the Internet have a UUCP link with an Internet
connected site such as UUNET. MX records are then provided so that electronic mail can
be sent to the site using the standard user@host notation. For example, a fictitious
domain foo.com might have the following MX records:

sun % host -t mx foo.com
foo.com MX relayl.UU.NET
foo.com MX relay2.UH.NET

MX records are used by mailers on hosts connected to the Internet. In this example the
other mailers are told "if you have mail to send to user@foo.com, send the mail to
relay1.uu.net or relay2.uu.net."

MX records have 16-bit integers assigned to them, called preference values. If multiple
MX MX records exist for a destination, they’re used in order, starting with the smallest
preference value.

Another example of MX records handles the case when a host is down or unavailable. In
that case the mailer uses the MX records only if it can’t connect to the destination using
TCP. In the case of the author’s primary system, which is connected to the In ternet by a
SLIP connection, which is down most of the time, we have:

sun % host -tv mx sun
Query about sun for record types MX
Trying sun within tuc.noao.edu ...
Query done, 2 answers, authoritative status: no error
sun.tuc.noao.edu 86400 IN MX 0 sun.tuc.noao.edu
sun.tuc.noao.edu 86400 IN MX 10 noao.edu

We also specified the -v option, to see the preference values. (This option also causes
other fields to be output.) The second field, 86400, is the time-to-live value in seconds.
This TTL is 24 hours (24 x 60 x 60). The third column, IN, is the class (Internet). We see
that direct delivery to the host itself, the first MX record, has the lowest preference value
of 0. If that doesn’t work (i.e., the SLIP link is down), the next higher preference is used
(10) and delivery is attempted to the host noao.edu. If that doesn’t work, the sender
will time out and retry at a later time.

In Section 28.3 we show examples of SMTP mail delivery using MX records.
Name server record. These specify the authoritative name server for a domain. They are
NS represented as domain names (a sequence of labels). We’ll see examples of these records
in the next section.

199

These are the common types of RRs. We’ll encounter many of them in later examples.

14.7 Caching
To reduce the DNS traffic on the Internet, all name servers employ a cache. With the standard Unix
implementation, the cache is maintained in the server, not the resolver. Since the resolver is part of
each application, and applications come and go, putting the cache into the program that lives the entire
time the system is up (the name server) makes sense. This makes the cache available to any
applications that use the server. Any other hosts at the site that use this name server also share the
server’s cache.

In the scenario that we’ve used for our examples so far (Figure 14.9), we’ve run the clients on the host
sun accessing the name server across the SLIP link on the host noao.edu. We’ll change that now
and run the name server on the host sun. In this way if we monitor the DNS traffic on the SLIP link
using tcpdump, we’ll only see queries that can’t be handled by the server out of its cache.

By default, the resolver looks for a name server on the local host (UDP port 53 or TCP port 53). We
delete the nameserver directive from our resolver file, leaving only the domain directive:

sun % cat /etc/resolv.conf
domain tuc.noao.edu

The absence of a nameserver directive in this file causes the resolver to use the name server on the
local host.

We then use the host command to execute the following query:

sun % host ftp.uu.net
ftp.uu.net A 192.48.96.9

Figure 14.14 shows the tcpdump output for this query.

sun.tuc.noao.edu.domain >
1 0.0 NS.NIC.DDN.MIL.domain:
2 A? ftp.uu.net. (28)
NS.NIC.DDN.MIL.domain >
0.559285 (
2 sun.tuc.noao.edu.domain:
0.5593)
2- 0/5/5 (229)
0.564449 ( sun.tuc.noao.edu.domain > ns.UU.NET.domain:
3
0.0052) 3+ A? ftp.uu.net. (28)
1.009476 ( ns.UU.NET.domain > sun.tuc.noao.edu.domain:
4
0.4450) 3* 1/0/0 A ftp.UU.NET (44)

Figure 14.14 tcpdump output for: host ftp.uu.net.

This time we’ve used a new option for tcpdump. We collected all the data to or from UDP or TCP
ports 53 with the -w option. This saves the raw output in a file for later processing. This prevents
tcpdump from trying to call the resolver itself, to print all the names corresponding to the IP
addresses. After we ran our queries, we terminated tcpdump and reran it with the -r option. This

200

causes it to read the raw output file and generate its normal printed output (which we show in Figure
14.14). This takes a few seconds, since tcpdump calls the resolver itself.

The first thing to notice in our tcpdump output is that the identifiers are small integers (2 and 3). This
is because we terminated the name server, and then restarted it, to force the cache to be empty. When
the name server starts up, it initializes the identifier to 1.

When we type our query, looking for the IP address of the host ftp.uu.net, the name server
contacts one of the eight root servers, ns.nic.ddn.mil (line 1). This is the normal A type query
that we’ve seen before, but notice that the recursion-desired flag is not specified. (A plus sign would
have been printed after the identifier 2 if the flag was set.) In our earlier examples we always saw the
resolver set the recursion-desired flag, but here we see that our name server doesn’t set the flag when
it’s contacting one of the root servers. This is because the root servers shouldn’t be asked to
recursively answer queries-they should be used only to find the addresses of other, authoritative
servers.

Line 2 shows that the response comes back with no answer RRs, five authority RRs, and five
additional information RRs. The minus sign following the identifier 2 means the recursion-available
(RA) flag was not set-this root server wouldn’t answer a recursive query even if we asked it to.

Although tcpdump doesn’t print the 10 RRs that are returned, we can execute the host command to
see what’s in the cache:

201

sun % host -v ftp.uu.net
Query about ftp.uu.net for record types A
Trying ftp.uu.net ...
Query done, 1 answer, status: no error
The following answer is not authoritative:
ftp.uu.net 19109 IN A 192.48.96.9
Authoritative nameservers:
UU.NET 170308 IN NS NS.UU.NET
UU.NET 170308 IN NS UUNET.UU.NET
UU.NET 170308 IN NS UUCP-GW-1.PA.DEC.COM
UU.NET 170308 IN NS UUCP-GW-2.PA.DEC.COM
UU.NET 170308 IN NS NS.EU.NET
Additional information:
NS.UU.NET 170347 IN A 137.39.1.3
UUNET.UU.NET 170347 IN A 192.48.96.2
UUCP-GW-1.PA.DEC.COM 170347 IN A 16.1.0.18
UUCP-GW-2.PA.DEC.COM 170347 IN A 16.1.0.19
NS.EU.NET 170347 IN A 192.16.202.11

This time we specified the -v option to see more than just the A record. This shows that there are five
authoritative name servers for the domain uu.net. The five RRs with additional information that are
returned by the root server contain the IP addresses of these five name servers. This saves us from
having to contact the root server again, to look up the address of one of the servers. This is another
implementation optimization in the DNS.

The host command states that the answer is not authoritative. This is because the answer was
obtained from our name server’s cache, not by contacting an authoritative server.

Returning to line 3 of Figure 14.14, our name server contacts the first of the authoritative servers
(ns.uu.net) with the same question: What is the IP address of ftp.uu.net? This time our server
sets the recursion-desired flag. The answer is returned on line 4 as a response with one answer RR.

We then execute the host command again, asking for the same name:

sun % host ftp.uu.net
ftp.uu.net A 192.48.96.9

202

This time there is no tcpdump output. This is what we expect, since the answer output by host is
returned from the server’s cache.

We execute the host command again, looking for the address of ftp.ee.lbl.gov:

sun%hostftp.ee.lbl.gov
ftp.ee.lbl.gov CNAME ee.lbl.gov
ee.lbl.gov A 128.3.112.20

Figure 14.15 shows the tcpdump output.

18.664971
1 c.nyser.net.domain:
(17.6555)
4 A? ftp.ee.lbl.gov. (32)
c.nyser.net.domain >
19.429412 (
0.7644)
4 0/4/4 (188)
19.432271 (
3 nsl.lbl.gov.domain:
0.0029)
5+ A? ftp.ee.lbl.gov. (32)
nsl.lbl.gov.domain >
19.909242 (
0.4770)
5* 2/0/0 CNAME ee.lbl.gov. (72)

Figure 14.15 tcpdump output for: host ftp.ee.lbl.gov.

Line 1 shows that this time our server contacts another of the root servers (c.nyser.net). A name
server normally cycles through the various servers for a zone until round-trip estimates are
accumulated. The server with the smallest round-trip time is then used.

Since our server is contacting a root server, the recursion-desired flag is not set. This root server does
not clear the recursion-available flag, as we saw in line 2 in Figure 14.14. (Even so, a name server still
should not ask a root server for a recursive query.)

In line 2 the response comes back with no answers, but four authority RRs and four additional
information RRs. As we can guess, the four authority RRs are the names of the name servers for
ftp.ee.lbl.gov, and the four other RRs contain the IP addresses of these four servers.

Line 3 is the query of the name server nsl.lbl.gov (the first of the four name servers returned in
line 2). The recursion-desired flag is set.

The response in line 4 is different from previous responses. Two answer RRs are returned and
tcpdump says that the first one is a CNAME RR. The canonical name of ftp.ee.lbl.gov is
ee.lbl.gov.

203

This is a common usage of CNAME records. The FTP site for LBL always has a name beginning with
ftp, but it may move from one host to another over time. Users need only know the name
ftp.ee.lbl.gov and the DNS will replace this with its canonical name when referenced.

Remember that when we ran host, it printed both the CNAME and the IP address of the canonical
name. This is because the response (line 4 in Figure 14.15) contained two answer RRs. The first one is
the CNAME and the second is the A record. If the A record had not been returned with the CNAME,
our server would have issued another query, asking for the IP address of ee.lbl.gov. This is
another implementation optimization-both the CNAME and the A record of the canonical name are
returned in one response.

14.8 UDP or TCP
We’ve mentioned that the well-known port numbers for DNS name servers are UDP port 53 and TCP
port 53. This implies that the DNS supports both UDP and TCP. But all the examples that we’ve
watched with tcpdump have used UDP. When is each protocol used and why?

When the resolver issues a query and the response comes back with the TC bit set ("truncated") it
means the size of the response exceeded 512 bytes, so only the first 512 bytes were returned by the
server. The resolver normally issues the request again, using TCP. This allows more than 512 bytes to
be returned. (Recall our discussion of the maximum UDP datagram size in Section 11.10.) Since TCP
breaks up a stream of user data into what it calls segments, it can transfer any amount of user data,
using multiple segments.

Also, when a secondary name server for a domain starts up it performs a zone transfer from the
primary name server for the domain. We also said that the secondary queries the primary on a regular
basis (often every 3 hours) to see if the primary has had its tables updated, and if so, a zone transfer is
performed. Zone transfers are done using TCP, since there is much more data to transfer than a single
query or response.

Since the DNS primarily uses UDP, both the resolver and the name server must perform their own
timeout and retransmission. Also, unlike many other Internet applications that used UDP (TFTP,
BOOTP, and SNMP), which operate mostly on local area networks, DNS queries and responses often
traverse wide area networks. The packet loss rate and variability in round-trip times are normally
higher on a WAN than a LAN, increasing the importance of a good retransmission and timeout
algorithm for DNS clients.

14.9 Another Example
Let’s look at another example that ties together many of the DNS features that we’ve described. We
start an Rlogin client, connecting to an Rlogin server in some other domain. Figure 14.16 shows the
exchange of packets that takes place.

204

Figure 14.16 Summary of packets exchanged to start up Rlogin client and server

The following 11 steps take place, assuming none of the information is already cached by the client or
server:

1. The client starts and calls its resolver function to convert the hostname that we typed into an IP
address. A query of type A is sent to a root server.
2. The root server’s response contains the name servers for the server’s domain.
3. The client’s resolver reissues the query of type A to the server’s name server. This query
normally has the recursion-desired flag set.
4. The response comes back with the IP address of the server host.
5. The Rlogin client establishes a TCP connection with the Rlogin server. (Chapter 18 provides all
the details of this step.) Three packets are exchanged between the client and server TCP modules.
6. The Rlogin server receives the connection from the client and calls its resolver to obtain the name
of the client host, given the IP address that the server receives from its TCP. This is a PTR query
issued to a root name server. This root server can be different from the root server used by the
client in step 1.
7. The root server’s response contains the name servers for the client’s in-addr.arpa domain.
8. The server’s resolver reissues the PTR query to the client’s name server.
9. The PTR response contains the FQDN of the client host.
10. The server’s resolver issues a query of type A to the client’s name server, asking for the IP
addresses corresponding to the name returned in the previous step. This may be done
automatically by the server’s gethostbyaddr function, as we described in Section 14.5,
otherwise the Rlogin server does this step explicitly. Also, the client’s name server is often the
same as the client’s in-addr.arpa name server, but this isn’t required.
11. The response from the client’s name server contains the A records for the client host. The Rlogin
server compares the A records with the IP address from. the client’s TCP connection request.

Caching can reduce the number of packets exchanged in this figure.

205

14.10 Summary
The DNS is an essential part of any host connected to the Internet, and widely used in private internets
also. The basic organization is a hierarchical tree that forms the DNS name space.

Applications contact resolvers to convert a hostname to an IP address, and vice versa. Resolvers then
contact a local name server, and this server may contact one of the root servers or other servers to
fulfill the request.

All DNS queries and responses have the same message format. This message contains questions and
possibly answer resource records (RRs), authority RRs, and additional RRs. We saw numerous
examples, showing the resolver configuration file and some of the DNS optimizations: pointers to
domain names (to reduce the size of messages), caching, the in-addr.arpa domain (to look up a
name given an IP address), and returning additional RRs (to save the requestor from issuing another
query).

Exercises
14.1 Classify a DNS resolver and a DNS name server as either client, server, or both.

14.2 Account for all 75 bytes in the response in Figure 14.12.

14.3 In Section 12.3 we said that an application that accepts either a dotted-decimal IP address or a
hostname should assume the former, and if that fails, then assume a hostname. What happens if the
order of the tests is reversed?

14.4 Every UDP datagram has an associated length. A process that receives a UDP datagram is told
what its length is. When a resolver issues a query using TCP instead of UDP, since TCP is a stream of
bytes without any record markers, how does the application know how much data is returned? Notice
that there is no length field in the DNS header (Figure 14.3). (Hint: Look at RFC 1035.)

14.5 We said that a name server must know the IP addresses of the root servers and that this
information is available via anonymous FTP. Unfortunately not all system administrators update their
DNS files whenever changes are made to the list of root servers. (Changes do occur to the list of root
servers, but not frequently.) How do you think the DNS handles this?

14.6 Fetch the file specified in Exercise 1.8 and determine who is responsible for maintaining the root
name servers. How frequently are the root servers updated?

14.7 What is a problem with maintaining the cache in the name server, and having a stateless resolver?

14.8 In the discussion of Figure 14.10 we said that the name server sorts the A records so that
addresses on common networks appear first. Who should sort the A records, the name server or the
resolver?

206

TFTP: Trivial File Transfer Protocol
15.1 Introduction
TFTP is the Trivial File Transfer Protocol. It is intended to be used when bootstrapping diskless
systems (normally workstations or X terminals). Unlike the File Transfer Protocol (FTP), which we
describe in Chapter 27 and which uses TCP, TFTP was designed to use UDP, to make it simple and
small. Implementations of TFTP (and its required UDP, IP, and a device driver) can fit in read-only
memory.

This chapter provides an overview of TFTP because we’ll encounter it in the next chapter with the
Bootstrap Protocol. We also encountered TFTP in Figure 5.1 when we bootstrapped the host sun from
the network. It issued a TFTP request after obtaining its IP address using RARP.

RFC 1350 [Sollins 1992] is the official specification of version 2 of TFTP. Chapter 12 of [Stevens
1990] provides a complete source code implementation of a TFTP client and server, and describes
some of the programming techniques used with TFTP.

15.2 Protocol
Each exchange between a client and server starts with the client asking the server to either read a file
for the client or write a file for the client. In the normal case of bootstrapping a diskless system, the
first request is a read request (RRQ). Figure 15.1 shows the format of the five TFTP messages.
(Opcodes 1 and 2 share the same format.)

Figure 15.1 Format of the five TFTP messages.

The first 2 bytes of the TFTP message are an opcode. For a read request (RRQ) and write request
(WRQ) the filename specifies the file on the server that the client wants to read from. or write to. We
specifically show that this filename is terminated by a byte of 0 in Figure 15.1. The mode is one of the
ASCII strings netascii or octet (in any combination of uppercase or lowercase), again
terminated by a byte of 0. netascii means the data are lines of ASCII text with each line terminated

207

by the 2-character sequence of a carriage return followed by a linefeed (called CR/LF). Both ends
must convert between this format and whatever the local host uses as a line delimiter. An octet
transfer treats the data as 8-bit bytes with no interpretation.

Each data packet contains a block number that is later used in an acknowledgment packet. As an
example, when reading a file the client sends a read request (RRQ) specifying the filename and mode.
If the file can be read by the client, the server responds with a data packet with a block number of 1.
The client sends an ACK of block number 1. The server responds with the next data packet, with a
block number of 2. The client sends an ACK of block number 2. This continues until the file is
transferred. Each data packet contains 512 bytes of data, except for the final packet, which contains
0-511 bytes of data. When the client receives a data packet with less than 512 bytes of data, it knows it
has received the final packet.

In the case of a write request (WRQ), the client sends the WRQ specifying the filename and mode. If
the file can be written by the client, the server responds with an ACK of block number 0. The client
then sends the first 512 bytes of file with a block number of 1. The server responds with an ACK of
block number 1.

This type of data transmission is called a stop-and-wait protocol. It is found only in simple protocols
such as TFTP. We’ll see in Section 20.3 that TCP provides a different form of acknowledgment,
which can provide higher throughput. TFTP is designed for simplicity of implementation, not high
throughput.

The final TFTP message type is the error message, with an opcode of 5. This is what the server
responds with if a read request or write request can’t be processed. Read and write errors during file
transmission also cause this message to be sent, and transmission is then terminated. The error number
gives a numeric error code, followed by an ASCII error message that might contain additional,
operating system specific information.

Since TFTP uses the unreliable UDP, it is up to TFTP to handle lost and duplicated packets. Lost
packets are detected with a timeout and retransmission implemented by the sender. (Be aware of a
potential problem called the "sorcerer’s apprentice syndrome" that can occur if both sides time out and
retransmit. Section 12.2 of [Stevens 1990] shows how the problem can occur.) As with most UDP
applications, there is no checksum in the TFTP messages, which assumes any corruption of the data
will be caught by the UDP checksum (Section 11.3).

15.3 An Example
Let’s examine TFTP by watching the protocol in action. We’ll run the TFTP client on the host bsdi
and fetch a text file from the host svr4:

208

bsdi % tftp svr4 start the TFTP client
tftp> get testl.c fetch a file from the server
Received 962 bytes in 0.3 seconds
tftp> quit and terminate
bsdi % ls -1 testl.c how many bytes in thefile we fetched?
-rw-r-r- 1 rstevens staff 914 Mar 20 11:41 test1.c
bsdi % wc -1 testl.c and how many lines?
48 testl.c

The first point that catches our eye is that the file contains 914 bytes under Unix, but TFTP transfers
962 bytes. Using the wc program we see that there are 48 lines in the file, so the 48 Unix newline
characters are expanded into 48 CR/LF pairs, since the TFTP default is a netascii transfer. Figure
15.2 shows the packet exchange that takes place.

1 0.0 bsdi.ll06 > svr4.tftp: 19 RRQ "testl.c"
2 0.287080 (0.2871) svr4.1077 > bsdi.ll06: udp 516
3 0.291178 (0.0041) bsdi.1106 > svr4.1077: udp 4
4 0.299446 (0.0083) svr4.1077 > bsdi.1106: udp 454
5 0.312320 (0.0129) bsdi.ll06 > svr4.1077: udp 4

Figure 15.2 Packet exchange for TFTP of a file.

Line 1 shows the read request from the client to the server. Since the destination UDP port is the TFTP
well-known port (69), tcpdump interprets the TFTP packet and prints RRQ and the name of the file.
The length of the UDP data is printed as 19 bytes and is accounted for as follows: 2 bytes for the
opcode, 7 bytes for the filename, 1 byte of 0, 8 bytes for netascii, and another byte of 0.

The next packet is from the server (line 2) and contains 516 bytes: 2 bytes for the opcode, 2 bytes for
the block number, and 512 bytes of data. Line 3 is the acknowledgment for this data: 2 bytes for the
opcode and 2 bytes for the block number.

The final data packet (line 4) contains 450 bytes of data. The 512 bytes of data in line 2 and this 450
bytes of data account for the 962 bytes of data output by the client.

Note that tcpdump doesn’t output any additional TFTP protocol information for lines 2-5, whereas it
interpreted the TFTP message in line 1. This is because the server’s port number changes between
lines 1 and 2. The TFTP protocol requires that the client send the first packet (the RRQ or WRQ) to
the server’s well-known UDP port (69). The server then allocates some other unused ephemeral port
on the server’s host (1077 in Figure 15.2), which is then used by the server for all further packet
exchange between this client and server. The client’s port number (1106 in this example) doesn’t
change, tcpdump has no idea that port 1077 on host svr4 is really a TFTP server.

209

The reason the server’s port number changes is so the server doesn’t tie up the well-known port for the
amount of time required to transfer the file (which could be many seconds or even minutes). Instead,
the well-known port is left available for other TFTP clients to send their requests to, while the current
transfer is under way.

Recall from Figure 10.6 that when the RIP server had more than 512 bytes to send to the client, both
UDP datagrams came from the server’s well-known port. In that example, even though the server had
to write multiple datagrams to send all the data back, the server did one write, followed by the next,
both from its well-known port. Here, with TFTP, the protocol is different since there is a longer term
relationship between the client and server (which we said could be seconds or minutes). If one server
process used the well-known port for the duration of the file transfer, it would either have to refuse any
further requests that arrived from other clients, or that one server process would have to multiplex file
transfers with multiple clients at the same time, on the same port (69). The simplest solution is to have
the server obtain a new port after it receives the RRQ or WRQ. Naturally the client must detect this
new port when it receives the first data packet (line 2 in Figure 15.2) and then send all further
acknowledgments (lines 3 and 5) to that new port. In Section 16.3 we’ll see TFTP used when an X
terminal is bootstrapped.

15.4 Security
Notice in the TFTP packets (Figure 15.1) that there is no provision for a username or password. This is
a feature (i.e., "security hole") of TFTP. Since TFTP was designed for use during the bootstrap process
it could be impossible to provide a username and password.

This feature of TFTP was used by many crackers to obtain copies of a Unix password file and then try
to guess passwords. To prevent this type of access, most TFTP servers nowadays provide an option
whereby only files in a specific directory (often /tftpboot on Unix systems) can be accessed. This
directory then contains only the bootstrap files required by the diskless systems.

For additional security, the TFTP server on a Unix system normally sets its user ID and group ID to
values that should not be assigned to any real user. This allows access only to files that have
world-read or world-write permissions.

15.5 Summary
TFTP is a simple protocol designed to fit into read-only memory and be used only during the bootstrap
process of diskless systems. It uses only a few message formats and a stop-and-wait protocol.

To allow multiple clients to bootstrap at the same time, a TFTP server needs to provide some form of
concurrency. Because UDP does not provide a unique connection between a client and server (as does
TCP), the TFTP server provides concurrency by creating a new UDP port for each client. This allows
different client input datagrams to be demultiplexed by the server’s UDP module, based on destination
port numbers, instead of doing this in the server itself.

The TFTP protocol provides no security features. Most implementations count on the system
administrator of the TFTP server to restrict any client’s access to the files necessary for bootstrapping
only.

Chapter 27 covers the File Transfer Protocol (FTP), which is designed for general purpose,
high-throughput file transfer.

210

Exercises
15.1 Read the Host Requirements RFC to see what a TFTP server should do if it receives a request and
the destination IP address of the request is a broadcast address.

15.2 What do you think happens when the TFTP block number wraps around from 65535 to 0? Does
RFC 1350 say anything about this?

15.3 We said that the TFTP sender performs the timeout and retransmission to handle lost packets.
How does this affect the use of TFTP when it’s being used as part of the bootstrap process?

15.4 What is the limiting factor in the time required to transfer a file using TFTP?

211

BOOTP: Bootstrap Protocol
16.1 Introduction
In Chapter 5 we described how a diskless system, with no knowledge of its IP address, can determine
its IP address using RARP when it is bootstrapped. There are two problems with RARP: (1) the only
thing returned is the IP address, and (2) since RARP uses a link-layer broadcast, RARP requests are
not forwarded by routers (necessitating an RARP server on every physical network). This chapter
describes an alternative method for a diskless system to bootstrap itself, called the Bootstrap Protocol,
or BOOTP.

BOOTP uses UDP and normally works in conjunction with TFTP (Chapter 15). RFC 951 [Croft and
Gilmore 1985] is the official specification for BOOTP with clarifications given in RFC 1542 [Wimer
1993].

16.2 BOOTP Packet Format
BOOTP requests and replies are encapsulated in UDP datagrams, as shown in Figure 16.1.

Figure 16.1 Encapsulation of BOOTP requests and replies within a UDP datagram.

Figure 16.2 shows the format of the 300-byte BOOTP request and reply.

212

Figure 16.2 Format of BOOTP request and reply.

Opcode is 1 for a request and 2 for a reply. The hardware type field is 1 for a 10 Mbits/sec Ethernet,
the same value that is in the field of the same name in an ARP request or reply (Figure 4.3). Similarly,
the hardware address length is 6 bytes for an Ethernet.

The hop count is set to 0 by the client, but can be used by a proxy server (described in Section 16.5).

The transaction ID is a 32-bit integer set by the client and returned by the server. This lets the client
match a response with a request. The client should set this to a random number for each request.

Number of seconds can be set by the client to the time since it started trying to bootstrap. The servers
can look at this value, and perhaps a secondary server for a client won’t respond until the number of
seconds has exceeded some value, implying that the client’s primary server is down.

213

If the client already knows its IP address, it fills in the client IP address. Otherwise, the client sets this
to 0. In the latter case the server fills in your IP address with the client’s IP address. The server IP
address is filled in by the server. If a proxy server is used (Section 16.5), that proxy server fills in its
gateway IP address.

The client must set its client hardware address. Although this is the same value as in the Ethernet
header, by placing the field in the UDP datagram also, it is easily available to any user process (e.g., a
BOOTP server) that receives the datagram. It is normally much harder (or impossible) for a process
reading UDP datagrams to determine the fields in the Ethernet header that carried the UDP datagram.

The server hostname is a null terminated string that is optionally filled in by the server. The server can
also fill in the boot filename with the fully qualified, null terminated pathname of a file to bootstrap
from.

The vendor-specific area is used for various extensions to BOOTP. Section 16.6 describes some of
these extensions.

When a client is bootstrapping using BOOTP (an opcode of 1) the request is normally a link-layer
broadcast and the destination IP address in the IP header is normally 255.255.255.255 (the limited
broadcast. Section 12.2). The source IP address is often 0.0.0.0 since the client does not know its own
IP address yet. Recall from Figure 3.9 that 0.0.0.0 is a valid source IP address when a system is
bootstrapping itself.

Port Numbers
There are two well-known ports for BOOTP: 67 for the server and 68 for the client. This means the
client does not choose an unused ephemeral port, but uses 68 instead. The reason two port numbers
were chosen, instead of just one for the server, is that a server’s reply can be (but normally isn’t)
broadcast.

If the server’s reply were broadcast, and if the client were to choose an ephemeral port number, these
broadcasts would also be received by other applications on other hosts that happen to be using the
same ephemeral port number. Hence, it is considered bad form to broadcast to a random (i.e.,
ephemeral) port number.

If the client also used the server’s well-known port (67) as its port, then all servers on the network are
awakened to look at each broadcast reply. (If all the servers were awakened, they would examine the
opcode, see that it’s a reply and not a request, and go back to sleep.) Therefore the choice was made to
have all clients use a single well-known port that differs from the server’s well-known port.

If multiple clients are bootstrapping at the same time, and if the server broadcasts the replies, each
client sees the replies intended for the other clients. The clients can use the transaction ID field in the
BOOTP header to match replies with requests, or the client can examine the returned client hardware
address.

16.3 An Example
Let’s look at an example of BOOTP when an X terminal is bootstrapped. Figure 16.3 shows the
tcpdump output. (The client’s name is proteus and the server’s name is mercury. This tcpdump
output was obtained on a different network from the one we’ve been using for all the other examples
in the text.)

214

0.0.0.0.68 > 255.255.255.255.bootp:
1 0.0
secs: 100 ether 0:0:a7:0:62:7c
mercury.bootp > proteus.68: secs:100
0.355446 Y:proteus
2
(0.3554) S: mercury G: mercury ether 0:0:a7:0:62:7c
file "/local/var/bootfiles/Xncdl9r"
0.355447
3 arp who-has proteus tell 0.0.0.0
(0.0000)
0.851508
4 arp who-has proteus tell 0.0.0.0
(0.4961)
1.371070
5 arp who-has proteus tell proteus
(0..5196)
1.863226 proteus.68 > 255.255.255.255.bootp:
6
(0..4922) secs: 100 ether 0:0:a7:0:62:7c
mercury.bootp > proteus.68: secsilOO
1.71038 Y:proteus
7
(0..0078) S: mercury G: mercury ether 0:0:a7:0;62:7c
3.871038 proteus.68 > 255.255.255.255.bootp:
8
(2.0000) secs: 1OO ether 0:0:a7:0:62:7c
mercury.bootp > proteus. 68: secs: 100
3.878850 Y:proteus
9
(0.0078) S: mercury G: mercury ether 0:0:a7:0:62:7c
5.925786
10 arp who-has mercury tell proteus
(2.0469)
5.929692
11 arp reply mercury is-at 8:0:2b:28:eb:1d
(0.0039)
5.929694 proteus. tftp > mercury, tftp: 37 RRQ
12
(0.0000) "/local/var/bootfiles/Xncdl9r"
5.996094 mercury.2352 > proteus.tftp: 516 DATA block
13
(0.0664) 1
6.000000
14 proteus. tftp > mercury. 2352: 4 ACK
(0.0039)
many lines deleted here
14.980472 mercury.2352 > proteus.tftp: 516 DATA block
15
(8.9805) 2510
14.984376
16 proteus. tftp > mercury. 2352: 4 ACK
(0.0039)

215

14.984377 mercury. 2352 > proteus. tftp: 228 DATA
17
(0.0000) block 2464
14.984378
18 proteus.tftp > mercury.2352: 4 ACK
(0.0000)

Figure 16.3 Example of BOOTP being used to bootstrap an X terminal.

In line 1 we see the client’s request from 0.0.0.0.68, destined for 255.255.255.255.67. The only fields
the client has filled in are the number of seconds and its Ethernet address. We’ll see that this client
always sets the number of seconds to 100. The hop count and transaction ID are both 0 since they are
not output by tcpdump. (A transaction ID of 0 means the client ignores the field, since it would set
this field to a random number if it was going to verify the returned value in the response.)

Line 2 is the reply from the server. The fields filled in by the server are the client’s IP address (which
tcpdump prints as the name proteus), the server’s IP address (printed as the name mercury), the
IP address of a gateway (printed as the name mercury), and the name of a boot file.

After receiving the BOOTP reply, the client immediately issues an ARP request to see if anyone else
on the network has its IP address. The name proteus following who-has corresponds to the target IP
address (Figure 4.3), and the sender’s IP address is set to 0.0.0.0. It sends another identical ARP
request 0.5 second later, and another one 0.5 second after that. In the third ARP request (line 5) it
changes the sender’s IP address to be its own IP address. This is a gratuitous ARP request (Section
4.7).

Line 6 shows that the client waits another 0.5 second and broadcasts another BOOTP request. The
only difference between this request and line 1 is that now the client puts its own IP address in the IP
header. It receives the same reply from the same server (line 7). The client waits another 2 seconds and
broadcasts yet another BOOTP request (line 8) and receives the same reply from the same server.

The client then waits another 2 seconds and sends an ARP request for its server mercury (line 10). The
ARP reply is received and the client immediately issues a TFTP read request for its boot file (line 12).
What follows are 2464 TFTP data packets and acknowledgments. The amount of data transferred is
512 x 2463+224 = 1,261,280 bytes. This loads the operating system into the X terminal. We have
deleted most of the TFTP lines from Figure 16.3.

One thing to notice, when comparing this TFTP exchange with Figure 15.2, is that here the client uses
the TFTP well-known port (69) for the entire transfer. Since one of the two partners is using port 69,
tcpdump knows that the packets are TFTP messages, so it is able to interpret each packet using the
TFTP protocol. This is why Figure 16.3 indicates which packets contain data, which contain
acknowledgments, and what the block number is for each packet. We didn’t get this additional
information in Figure 15.2 because neither end was using TFTP’s well-known port for the data
transfer. Normally the TFTP client cannot use TFTP’s well-known port, since that port is used by the
server on a multiuser system. But here the system is being bootstrapped, so a TFTP server is not
provided, allowing the client to use the port for the duration of the transfer. This also implies that the
TFTP server on mercury doesn’t care what the client’s port number is-it sends the data to the client’s
port, whatever that happens to be.

From Figure 16.3 we see that 1,261,280 bytes are transferred in 9 seconds. This is a rate of about
140,000 bytes per second. While this is slower than most FTP file transfers across an Ethernet, it is not
that bad for a simple stop-and-wait protocol such as TFTP.

216

What follows as this X terminal is bootstrapped are additional TFTP transfers of the terminal’s font
files, some DNS name server queries, and then the initialization of the X protocol. The total time in
Figure 16.3 was almost 15 seconds, and another 6 seconds is taken for the remaining steps. This gives
a total of 21 seconds to bootstrap the diskless X terminal.

16.4 BOOTP Server Design
The BOOTP client is normally provided in read-only memory on the diskless system. It is interesting
to see how the server is normally implemented.

First, the server reads UDP datagrams from its well-known port (67). Nothing special is required. This
differs from an RARP server (Section 5.4), which we said had to read Ethernet frames with a type field
of "RARP request." The BOOTP protocol also made it easy for the server to obtain the client’s
hardware address, by placing it into the BOOTP packet (Figure 16.2).

An interesting problem arises: how can the server send a response directly back to the client? The
response is a UDP datagram, and the server knows the client’s IP address (probably read from a
configuration file on the server). But if the client sends a UDP datagram to that IP address (the normal
way UDP output is handled), the server’s host will probably issue an ARP request for that IP address.
But the client can’t respond to the ARP request since it doesn’t know its IP address yet! (This is called
the "chicken and egg" issue in RFC 951.)

There are two solutions. The first, commonly used by Unix servers, is tor the server to issue an
ioctl(2) request to the kernel, to place an entry into the ARP cache for this client. (This is what the
arp -s command does. Section 4.8.) The server can do this since it knows the client’s hardware
address and IP address. This means that when the server sends the UDP datagram (the BOOTP reply),
the server’s ARP module will find the client’s IP address in the ARP cache.

An alternative solution is tor the server to broadcast the BOOTP reply, instead of sending it directly to
the client. Since reducing the number of broadcasts on a network is always desirable, this solution
should be used only if the server cannot make an entry into its ARP cache. Normally it requires
superuser permission to make an entry into the ARP cache, requiring a broadcast reply if the server is
nonprivileged.

16.5 BOOTP Through a Router
We said in Section 5.4 that one of the drawbacks of RARP is that it uses a link-layer broadcast, which
is normally not forwarded by a router. This required an RARP server on each physical network.
BOOTP can be used through a router, if supported by the router. (Most major router vendors do
support this feature.)

This is mainly intended for diskless routers, because if a multiuser system with a disk is used as a
router, it can probably run a BOOTP server itself. Alternatively, the common Unix BOOTP server
(Appendix F) supports this relay mode, but again, if you can run a BOOTP server on the physical
network, there’s normally no need to forward the requests to yet another server on another network.

What happens is that the router (also called the "BOOTP relay agent") listens tor BOOTP requests on
the server’s well-known port (67). When a request is received, the relay agent places its IP address into
the gateway IP address field in the BOOTP request, and sends the request to the real BOOTP server.
(The address placed by the relay agent into the gateway field is the IP address of the interface on
which the request was received.) The relay agent also increments the hops field by one. (This is to

217

prevent infinite loops in case the request is reforwarded. RFC 951 mentions that the request should
probably be thrown away if the hop count reaches 3.) Since the outgoing request is a unicast datagram
(as opposed to the original client request that was broadcast), it can follow any route to the real
BOOTP server, passing through other routers. The real server gets the request, forms the BOOTP
reply, and sends it back to the relay agent, not the client. The real server knows that the request has
been forwarded, since the gateway field in the request is nonzero. The relay agent receives the reply
and sends it to the client.

16.6 Vendor-Specific Information
In Figure 16.2 we showed a 64-byte vendor-specific area, RFC 1533 [Alexander and Droms 1993]
defines the format of this area. This area contains optional information for the server to return to the
client.

If information is provided, the first 4 bytes of this area are set to the IP address 99.130.83.99. This is
called the magic cookie and means there is information in the area.

The rest of this area is a list of items. Each item begins with a 1-byte tag field. Two of the items
consist of just the tag field: a tag of 0 is a pad byte (to force following items to preferred byte
boundaries), and a tag of 255 marks the end of the items. Any bytes remaining in this area after the
first end byte should be set to this value (255).

Other than these two 1-byte items, the remaining items consist of a single length byte, followed by the
information. Figure 16.4 shows the format of some of the items in the vendor-specific area.

Figure 16.4 Format of some of the items in the vendor-specific area.

218

The subnet mask and time value are really fixed-length items because their values always occupy 4
bytes. The time offset is the number of seconds since midnight January 1,1900, UTC.

The gateway item is an example of a variable-length item. The length is always a multiple of 4, and
the values are the 32-bit IP addresses of one or more gateways (routers) for the client to use. The first
one returned must be the preferred gateway.

There are 14 other items defined in RFC 1533. Probably the most important is the IP address of a DNS
name server, with a tag value of 6. Other items return the IP address of a printer server, the IP address
of a time server, and so on. Refer to the RFC for all the details.

Returning to our example in Figure 16.3, we never saw an ICMP address mask request (Section 6.3)
that would have been broadcast by the client to find its subnet mask. Although it wasn’t output by
tcpdump, we can probably assume that the client’s subnet mask was returned in the vendor-specific
area of the BOOTP reply.

The Host Requirements RFC recommends that a system using BOOTP obtain its subnet mask using BOOTP, not ICMP.

The size of the vendor-specific area is limited to 64 bytes. This is a constraint for some applications. A
new protocol named DHCP (Dynamic Host Configuration Protocol) is built on, but replaces, BOOTP.
DHCP extends this area to 312 bytes and is defined in RFC 1541 [Droms 1993].

16.7 Summary
BOOTP uses UDP and is intended as an alternative to RARP for bootstrapping a diskless system to
find its IP address. BOOTP can also return additional information, such as the IP address of a router,
the client’s subnet mask, and the IP address of a name server.

Since BOOTP is used in the bootstrap process, a diskless system needs the following protocols
implemented in read-only memory: BOOTP, TFTP, UDP, IP, and a device driver for the local
network.

The implementation of a BOOTP server is easier than an RARP server, since BOOTP requests and
replies are in UDP datagrams, not special link-layer frames. A router can also serve as a proxy agent
for a real BOOTP server, forwarding client requests to the real server on a different network.

Exercises
16.1 We’ve said that one advantage of BOOTP over RARP is that BOOTP can work through routers,
whereas RARP, which is a link-layer broadcast, cannot. Yet in Section 16.5 we had to define special
ways for BOOTP to work through a router. What would happen if a capability were added to routers
allowing them to forward RARP requests?

16.2 We said that a BOOTP client must use the transaction ID to match responses with requests, in
case there are multiple clients bootstrapping at the same time from a server that broadcasts replies. But
in Figure 16.3 the transaction ID is 0, implying that this client ignores the transaction ID. How do you
think this client matches the responses with its requests?

219

TCP: Transmission Control Protocol
17.1 Introduction
In this chapter we provide a description of the services provided by TCP for the application layer. We
also look at the fields in the TCP header. In the chapters that follow we examine all of these header
fields in more detail, as we see how TCP operates.

Our description of TCP starts in this chapter and continues in the next seven chapters. Chapter 18
describes how a TCP connection is established and terminated, and Chapters 19 and 20 look at the
normal transfer of data, both for interactive use (remote login) and bulk data (file transfer). Chapter 21
provides the details of TCP’s timeout and retransmission, followed by two other TCP timers in
Chapters 22 and 23. Finally Chapter 24 takes a look at newer TCP features and TCP performance.

The original specification for TCP is RFC 793 [Postel 1981c], although some errors in that RFC are
corrected in the Host Requirements RFC.

17.2 TCP Services
Even though TCP and UDP use the same network layer (IP), TCP provides a totally different service
to the application layer than UDP does. TCP provides a connection-oriented, reliable, byte stream
service.

The term connection-oriented means the two applications using TCP (normally considered a client and
a server) must establish a TCP connection with each other before they can exchange data. The typical
analogy is dialing a telephone number, waiting for the other party to answer the phone and say "hello,"
and then saying who’s calling. In Chapter 18 we look at how a connection is established, and
disconnected some time later when either end is done.

There are exactly two end points communicating with each other on a TCP connection. Concepts that
we talked about in Chapter 12, broadcasting and multicasting, aren’t applicable to TCP.

TCP provides reliability by doing the following:

The application data is broken into what TCP considers the best sized chunks to send. This is
totally different from UDP, where each write by the application generates a UDP datagram of that
size. The unit of information passed by TCP to IP is called a segment. (See Figure 1.7) In Section
18.4 we’ll see how TCP decides what this segment size is.
When TCP sends a segment it maintains a timer, waiting for the other end to acknowledge
reception of the segment. If an acknowledgment isn’t received in time, the segment is
retransmitted. In Chapter 21 we’ll look at TCP’s adaptive timeout and retransmission strategy.
When TCP receives data from the other end of the connection, it sends an acknowledgment. This
acknowledgment is not sent immediately, but normally delayed a fraction of a second, as we
discuss in Section 19.3.
TCP maintains a checksum on its header and data. This is an end-to-end checksum whose
purpose is to detect any modification of the data in transit. If a segment arrives with an invalid
checksum, TCP discards it and doesn’t acknowledge receiving it. (It expects the sender to time
out and retransmit.)
Since TCP segments are transmitted as IP datagrams, and since IP datagrams can arrive out of
order, TCP segments can arrive out of order. A receiving TCP resequences the data if necessary,

220

passing the received data in the correct order to the application.
Since IP datagrams can get duplicated, a receiving TCP must discard duplicate data.
TCP also provides flow control. Each end of a TCP connection has a finite amount of buffer
space. A receiving TCP only allows the other end to send as much data as the receiver has buffers for.
This prevents a fast host from taking all the buffers on a slower host.

A stream of 8-bit bytes is exchanged across the TCP connection between the two applications. There
are no record markers automatically inserted by TCP. This is what we called a byte stream service. If
the application on one end writes 10 bytes, followed by a write of 20 bytes, followed by a write of 50
bytes, the application at the other end of the connection cannot tell what size the individual writes
were. The other end may read the 80 bytes in four reads of 20 bytes at a time. One end puts a stream of
bytes into TCP and the same, identical stream of bytes appears at the other end.

Also, TCP does not interpret the contents of the bytes at all. ’TCP has no idea if the data bytes being
exchanged are binary data, ASCII characters, EBCDIC characters, or whatever. The interpretation of
this byte stream is up to the applications on each end of the connection.

This treatment of the byte stream by TCP is similar to the treatment of a file by the Unix operating system. The Unix kernel
does no interpretation whatsoever of the bytes that an application reads or write-that is up to the applications. There is no
distinction to the Unix kernel between a binary file or a file containing lines of text.

17.3 TCP Header
Recall that TCP data is encapsulated in an IP datagram, as shown in Figure 17.1.

Figure 17.1 Encapsulation of TCP data in an IP datagram.

221

Figure 17.2 shows the format of the TCP header. Its normal size is 20 bytes, unless options are
present.

Figure 17.2 TCP header.

Each TCP segment contains the source and destination port number to identify the sending and
receiving application. These two values, along with the source and destination IP addresses in the IP
header, uniquely identify each connection.

The combination of an IP address and a port number is sometimes called a socket. This term appeared
in the original TCP specification (RFC 793), and later it also became used as the name of the
Berkeley-derived programming interface (Section 1.15). It is the socket pair (the 4-tuple consisting of
the client IP address, client port number, server IP address, and server port number) that specifies the
two end points that uniquely identifies each TCP connection in an internet.

The sequence number identifies the byte in the stream of data from the sending TCP to the receiving
TCP that the first byte of data in this segment represents. If we consider the stream of bytes flowing in
one direction between two applications, TCP numbers each byte with a sequence number. This
sequence number is a 32-bit unsigned number that wraps back around to 0 after reaching 2 32 - 1.

When a new connection is being established, the SYN flag is turned on. The sequence number field
contains the initial sequence number (ISN) chosen by this host for this connection. The sequence
number of the first byte of data sent by this host will be the ISN plus one because the SYN flag
consumes a sequence number. (We describe additional details on exactly how a connection is
established and terminated in the next chapter where we’ll see that the FIN flag consumes a sequence
number also.)

222

Since every byte that is exchanged is numbered, the acknowledgment number contains the next
sequence number that the sender of the acknowledgment expects to receive. This is therefore the
sequence number plus 1 of the last successfully received byte of data. This field is valid only if the
ACK flag (described below) is on.

Sending an ACK costs nothing because the 32-bit acknowledgment number field is always part of the
header, as is the ACK flag. Therefore we’ll see that once a connection is established, this field is
always set and the ACK flag is always on.

TCP provides a full-duplex service to the application layer. This means that data can be flowing in
each direction, independent of the other direction. Therefore, each end of a connection must maintain a
sequence number of the data flowing in each direction.

TCP can be described as a sliding-window protocol without selective or negative acknowledgments.
(The sliding window protocol used for data transmission is described in Section 20.3.) We say that
TCP lacks selective acknowledgments because the acknowledgment number in the TCP header means
that the sender has successfully received up through but not including that byte. There is currently no
way to acknowledge selected pieces of the data stream. For example, if bytes 1-1024 are received OK,
and the next segment contains bytes 2049-3072, the receiver cannot acknowledge this new segment.
All it can send is an ACK with 1025 as the acknowledgment number. There is no means for negatively
acknowledging a segment. For example, if the segment with bytes 1025-2048 did arrive, but had a
checksum error, all the receiving TCP can send is an ACK with 1025 as the acknowledgment number.
In Section 21.7 we’ll see how duplicate acknowledgments can help determine that packets have been
lost.

The header length gives the length of the header in 32-bit words. This is required because the length of
the options field is variable. With a 4-bit field, TCP is limited to a 60-byte header. Without options,
however, the normal size is 20 bytes.

There are six flag bits in the TCP header. One or more of them can be turned on at the same time. We
briefly mention their use here and discuss each flag in more detail in later chapters.

URG The urgent pointer is valid (Section 20.8).
ACK The acknowledgment number is valid.
PSH The receiver should pass this data to the application as soon as possible (Section 20.5).
RST Reset the connection (Section 18.7).
Synchronize sequence numbers to initiate a connection. This flag and the next are
SYN
described in Chapter 18.
FIN The sender is finished sending data.

TCP’s flow control is provided by each end advertising a window size. This is the number of bytes,
starting with the one specified by the acknowledgment number field, that the receiver is willing to
accept. This is a 16-bit field, limiting the window to 65535 bytes. In Section 24.4 we’ll look at the new
window scale option that allows this value to be scaled, providing larger windows.

223

The checksum covers the TCP segment: the TCP header and the TCP data. This is a mandatory field
that must be calculated and stored by the sender, and then verified by the receiver. The TCP checksum
is calculated similar to the UDP checksum, using a pseudo-header as described in Section 11.3.

The urgent pointer is valid only if the URG flag is set. This pointer is a positive offset that must be
added to the sequence number field of the segment to yield the sequence number of the last byte of
urgent data. TCP’s urgent mode is a way for the sender to transmit emergency data to the other end.
We’ll look at this feature in Section 20.8.

The most common option field is the maximum segment size option, called the MSS. Each end of a
connection normally specifies this option on the first segment exchanged (the one with the SYN flag
set to establish the connection). It specifies the maximum sized segment that the sender wants to
receive. We describe the MSS option in more detail in Section 18.4, and some of the other TCP
options in Chapter 24.

In Figure 17.2 we note that the data portion of the TCP segment is optional. We’ll see in Chapter 18
that when a connection is established, and when a connection is terminated, segments are exchanged
that contain only the TCP header with possible options. A header without any data is also used to
acknowledge received data, if there is no data to be transmitted in that direction. There are also some
cases dealing with timeouts when a segment can be sent without any data.

17.4 Summary
TCP provides a reliable, connection-oriented, byte stream, transport layer service. We looked briefly at
all the fields in the TCP header and will examine them in detail in the following chapters.

TCP packetizes the user data into segments, sets a timeout any time it sends data, acknowledges data
received by the other end, reorders out-of-order data, discards duplicate data, provides end-to-end flow
control, and calculates and verifies a mandatory end-to-end checksum.

TCP is used by many of the popular applications, such as Telnet, Rlogin, FTP, and electronic mail
(SMTP).

Exercises
17.1 We’ve covered the following packet formats, each of which has a checksum in its corresponding
header: IP, ICMP, IGMP, UDP, and TCP. For each one, describe what portion of an IP datagram the
checksum covers and whether the checksum is mandatory or optional.

17.2 Why do all the Internet protocols that we’ve discussed (IP, ICMP, IGMP, UDP, TCP) quietly
discard a packet that arrives with a checksum error?

17.3 TCP provides a byte-stream service where record boundaries are not maintained between the
sender and receiver. How can applications provide their own record markers?

17.4 Why are the source and destination port numbers at the beginning of the TCP header?

17.5 Why does the TCP header have a header length field while the UDP header (Figure 11.2) does
not?

224

TCP Connection Establishment and Termination
18.1 Introduction
TCP is a connection-oriented protocol. Before either end can send data to the other, a connection must
be established between them. In this chapter we take a detailed look at how a TCP connection is
established and later terminated.

This establishment of a connection between the two ends differs from a connectionless protocol such
as UDP. We saw in Chapter 11 that with UDP one end just sends a datagram to the other end, without
any preliminary handshaking.

18.2 Connection Establishment and Termination
To see what happens when a TCP connection is established and then terminated, we type the following
command on the system svr4:

svr4 % telnet bsdi discard
Trying 192.82.148.3 ...
Connected to bsdi.
Escape character is ’^]’.
^] type Control, right bracket to talk to the Telnet client
telnet> quit terminate the connection
Connection closed.

The telnet command establishes a TCP connection with the host bsdi on the port corresponding to
the discard service (Section 1.12). This is exactly the type of service we need to see what happens
when a connection is established and terminated, without having the server initiate any data exchange.

tcpdump Output
Figure 18.1 shows the tcpdump output for the segments generated by this command.

225

svr4.1037 > bsdi.discard: S
1 0.0 1415531521:1415531521(0)
win 4096 <mss 1024>
bsdi.discard > svr4.1037: S
0.002402
2 1823083521:1823083521(0)
(0.0024)
ack 1415531522 win 4096 <mss 1024>
0.007224 svr4.1037 > bsdi.discard: ack 1823083522 win
3
(0.0048) 4096
svr4.1037 > bsdi.discard: F
4.155441
4 1415531522:1415531522(0)
(4.1482)
ack 1823083522 win 4096
4.156747 bsdi.discard > svr4.1037: . ack 1415531523
5
(0.0013) win 4096
bsdi.discard > svr4.1037: F
4.158144
6 1823083522:1823083522(0)
(0.0014)
ack 1415531523 win 4096
4.180662 svr4.1037 > bsdi.discard: . ack 1823083523
7
(0.0225) win 4096

Figure 18.1 tcpdump output for TCP connection establishment and termination.

These seven TCP segments contain TCP headers only. No data is exchanged. For TCP segments, each
output line begins with

source > destination: flags

where flags represents four of the six flag bits in the TCP header (Figure 17.2). Figure 18.2 shows the
five different characters that can appear in the flags output.

flag 3-character abbreviation Description
S SYN synchronize sequence numbers
F FIN sender is finished sending data
R RST reset connection
P PSH push data to receiving process as soon as possible
. - none of above four flags is on

Figure 18.2 flag characters output by tcpdump for flag bits in TCP header.

In this example we see the S, F, and period. We’ll see the other two flags (R and P) later. The other
two TCP header flag bits-ACK and URG-are printed specially by tcpdump.

It’s possible for more than one of the four flag bits in Figure 18.2 to be on in a single segment, but we
normally see only one on at a time.

226

RFC 1025 [Postel 1987], the TCP and IP Bake Off, calls a segment with the maximum combination of allowable flag bits
turned on at once (SYN, URG, PSH, FIN, and 1 byte of data) a Kamikaze packet. It’s also known as a nastygram,
Christmas tree packet, and lamp test segment.

In line 1, the field 1415531521:1415531521 (0) means the sequence number of the packet was
1415531521 and the number of data bytes in the segment was 0. tcpdump displays this by printing
the starting sequence number, a colon, the implied ending sequence number, and the number of data
bytes in parentheses. The advantage of displaying both the sequence number and the implied ending
sequence number is to see what the implied ending sequence number is, when the number of bytes is
greater than 0. This field is output only if (1) the segment contains one or more bytes of data or (2) the
SYN, FIN, or RST flag was on. Lines 1, 2, 4, and 6 in Figure 18.1 display this field because of the flag
bits-we never exchange any data in this example.

In line 2 the field ack 1415531522 shows the acknowledgment number. This is printed only if the
ACK flag in the header is on.

The field win 4096 in every line of output shows the window size being advertised by the sender. In
these examples, where we are not exchanging any data, the window size never changes from its default
of 4096. (We examine TCP’s window size in Section 20.4.)

The final field that is output in Figure 18.1, <mss 1024> shows the maximum segment size (MSS)
option specified by the sender. The sender does not want to receive TCP segments larger than this
value. This is normally to avoid fragmentation (Section 11.5). We discuss the maximum segment size
in Section 18.4, and show the format of the various TCP options in Section 18.10.

Time Line
Figure 18.3 shows the time line for this sequence of packets. (We described some general features of
these time lines when we showed the first one in Figure 6.11.) This figure shows which end is sending
packets. We also expand some of the tcpdump output (e.g., printing SYN instead of S). In this time
line we have also removed the window size values, since they add nothing to the discussion.

Connection Establishment Protocol
Now let’s return to the details of the TCP protocol that are shown in Figure 18.3. To establish a TCP
connection:

1. The requesting end (normally called the client) sends a SYN segment specifying the port number
of the server that the client wants to connect to, and the client’s initial sequence number (ISN,
1415531521 in this example). This is segment 1.
2. The server responds with its own SYN segment containing the server’s initial sequence number
(segment 2). The server also acknowledges the client’s SYN by ACKing the client’s ISN plus
one. A SYN consumes one sequence number.
3. The client must acknowledge this SYN from the server by ACKing the server’s ISN plus one
(segment 3).

These three segments complete the connection establishment. This is often called the three-way
handshake.

[IMAGE]
Figure 18.3 Time line of connection establishment and connection termination.

227

The side that sends the first SYN is said to perform an active open. The other side, which receives this
SYN and sends the next SYN, performs a passive open. (In Section 18.8 we describe a simultaneous
open where both sides can do an active open.)

When each end sends its SYN to establish the connection, it chooses an initial sequence number for
that connection. The ISN should change over time, so that each connection has a different ISN. RFC
793 [Postel 1981c] specifies that the ISN should be viewed as a 32-bit counter that increments by one
every 4 microseconds. The purpose in these sequence numbers is to prevent packets that get delayed in
the network from being delivered later and then misinterpreted as part of an existing connection.

How are the sequence numbers chosen? In 4.4BSD (and most Berkeley-derived implementations) when the system is
initialized the initial send sequence number is initialized to 1. This practice violates the Host Requirements RFC. (A
comment in the code acknowledges that this is wrong.) This variable is then incremented by 64,000 every half-second, and
will cycle back to 0 about every 9.5 hours. (This corresponds to a counter that is incremented every 8 microseconds, not
every 4 microseconds.) Additionally, each time a connection is established, this variable is incremented by 64,000.

The 4.1-second gap between segments 3 and 4 is the time between establishing the connection and
typing the quit command to telnet to terminate the connection.

Connection Termination Protocol
While it takes three segments to establish a connection, it takes four to terminate a connection. This is
caused by TCP’s half-close. Since a TCP connection is full-duplex (that is, data can be flowing in each
direction independently of the other direction), each direction must be shut down independently. The
rule is that either end can send a FIN when it is done sending data. When a TCP receives a FIN, it
must notify the application that the other end has terminated that direction of data flow. The sending of
a FIN is normally the result of the application issuing a close.

The receipt of a FIN only means there will be no more data flowing in that direction. A TCP can still
send data after receiving a FIN. While it’s possible for an application to take advantage of this
half-close, in practice few TCP applications use it. The normal scenario is what we show in Figure
18.3. We describe the half-close in more detail in Section 18.5.

We say that the end that first issues the close (e.g., sends the first FIN) performs the active close and
the other end (that receives this FIN) performs the passive close. Normally one end does the active
close and the other does the passive close, but we’ll see in Section 18.9 how both ends can do an active
close.

Segment 4 in Figure 18.3 initiates the termination of the connection and is sent when the Telnet client
closes its connection. This happens when we type quit. This causes the client TCP to send a FIN,
closing the flow of data from the client to the server.

When the server receives the FIN it sends back an ACK of the received sequence number plus one
(segment 5). A FIN consumes a sequence number, just like a SYN. At this point the server’s TCP also
delivers an end-of-file to the application (the discard server). The server then closes its connection,
causing its TCP to send a FIN (segment 6), which the client TCP must ACK by incrementing the
received sequence number by one (segment 7).

Figure 18.4 shows the typical sequence of segments that we’ve described for the termination of a
connection. We omit the sequence numbers. In this figure sending the FINs is caused by the
applications closing their end of the connection, whereas the ACKs of these FINs are automatically
generated by the TCP software.

228

Connections are normally initiated by the client, with the first SYN going from the client to the server.
Either end can actively close the connection (i.e., send the first FIN). Often, however, it is the client
that determines when the connection should be terminated, since client processes are often driven by
an interactive user, who enters something like "quit" to terminate. In Figure 18.4 we can switch the
labels at the top, calling the left side the server and the right side the client, and everything still works
fine as shown. (The first example in Section 14.4, for example, shows the daytime server closing the
connection.)

[IMAGE]
Figure 18.4 Normal exchange of segments during connection termination.

Normal tcpdump Output
Having to sort through all the huge sequence numbers is cumbersome, so the default tcpdump output
shows the complete sequence numbers only on the SYN segments, and shows all following sequence
numbers as relative offsets from the original sequence numbers. (To generate the output for Figure
18.1 we had to specify the -S option.) The normal tcpdump output corresponding to Figure 18.1 is

svr4.1037 > bsdi.discard: S
1 0.0 1415531521:1415531521(0)
win 4096 <mss 1024>
bsdi.discard > svr4.1037: S
0.002402
2 1823083521:1823083521(0)
(0.0024)
ack 1415531522 win 4096 <mss 1024>
0.007224
3 svr4.1037 > bsdi.discard: . ack 1 win 4096
(0.0048)
4.155441 svr4.1037 > bsdi.discard: F 1:1(0) ack 1 win
4
(4.1482) 4096
4.156747
5 bsdi.discard > svr4.1037: . ack 2 win 4096
(0.0013)
4.158144 bsdi.discard > svr4.1037: F 1:1(0) ack 2 win
6
(0.0014) 4096
4.180662
7 svr4.1037 > bsdi.discard: . ack 2 win 4096
(0.0225)

Figure 18.5 Normal tcpdump output for connection establishment and termination.

Unless we need to show the complete sequence numbers, we’ll use this form of output in all following
examples.

18.3 Timeout of Connection Establishment
There are several instances when the connection cannot be established. In one example the server host
is down. To simulate this scenario we issue our telnet command after disconnecting the Ethernet
cable from the server’s host. Figure 18.6 shows the tcpdump output.

229

bsdi-1024 > svr4.discard: S
1 0.0 291008001:291008001(0)
win 4096 <mss 1024> [tos 0x10]
bsdi-1024 > svr4.discard: S
5.814797 (
2 291008001:291008001(0)
5.8148)
win 4096 <mss 1024> [tos 0x10]
bsdi.l024 > svr4.discard: S
29.815436
3 291008001:291008001(0)
(24.0006)
win 4096 <mss 1024> [tos 0x10]

Figure 18.6 tcpdump output for connection establishment that times out.

The interesting point in this output is how frequently the client’s TCP sends a SYN to try to establish
the connection. The second segment is sent 5.8 seconds after the first, and the third is sent 24 seconds
after the second.

As a side note, this example was run about 38 minutes after the client was rebooted. This corresponds with the initial
sequence number of 291,008,001 (approximately 38 x 60 x 64000 x 2). Recall earlier in this chapter we said that typical
Berkeley-derived systems initialize the initial sequence number to 1 and then increment it by 64,000 every half-second.

Also, this is the first TCP connection since the system was bootstrapped, which is why the client’s port number is 1024.

What isn’t shown in Figure 18.6 is how long the client’s TCP keeps retransmitting before giving up.
To see this we have to time the telnet command:

bsdi % date ; telnet svr4 discard ; date
Thu Sep 24 16:24:11 MST 1992
Trying 192.82.148.2...
telnet: Unable to connect to remote host: Connection timed out
Thu Sep 24 16:25:27 MST 1992

The time difference is 76 seconds. Most Berkeley-derived systems set a time limit of 75 seconds on
the establishment of a new connection. We’ll see in Section 21.4 that the third packet sent by the client
would have timed out around 16:25:29, 48 seconds after it was sent, had the client not given up after
75 seconds.

First Timeout Period
One puzzling item in Figure 18.6 is that the first timeout period, 5.8 seconds, is close to 6 seconds, but
not exact, while the second period is almost exactly 24 seconds. Ten more of these tests were run and
the first timeout period took on various values between 5.59 seconds and 5.93 seconds. The second
timeout period, however, was always 24.00 (to two decimal places).

What’s happening here is that BSD implementations of TCP run a timer that goes off every 500 ms.
This 500-ms timer is used for various TCP timeouts, all of which we cover in later chapters. When we
type in the telnet command, an initial 6-second timer is established (12 clock ticks), but it may
expire anywhere between 5.5 and 6 seconds in the future. Figure 18.7 shows what’s happening.

230

[IMAGE]
Figure 18.7 TCP 500-ms timer.

Although the timer is initialized to 12 ticks, the first decrement of the timer can occur between 0 and
500 ms after it is set. From that point on the timer is decremented about every 500 ms, but the first
period can be variable. (We use the qualifier "about" because the time when TCP gets control every
500 ms can be preempted by other interrupts being handled by the kernel.)

When that 6-second timer expires at the tick labeled 0 in Figure 18.7, the timer is reset for 24 seconds
(48 ticks) in the future. This next timer will be close to 24 seconds, since it was set at a time when the
TCP’s 500-ms timer handler was called by the kernel.

Type-of-Service Field
In Figure 18.6, the notation [tos 0x10] appears. This is the type-of-service(TOS) field in the IP
datagram (Figure 3.2). The BSD/386 Telnet client sets the field for minimum delay.

18.4 Maximum Segment Size
The maximum segment size (MSS) is the largest "chunk" of data that TCP will send to the other end.
When a connection is established, each end can announce its MSS. The values we’ve seen have all
been 1024. The resulting IP datagram is normally 40 bytes larger: 20 bytes for the TCP header and 20
bytes for the IP header.

Some texts refer to this as a "negotiated" option. It is not negotiated in any way. When a connection is
established, each end has the option of announcing the MSS it expects to receive. (An MSS option can
only appear in a SYN segment.) If one end does not receive an MSS option from the other end, a
default of 536 bytes is assumed. (This default allows for a 20-byte IP header and a 20-byte TCP header
to fit into a 576-byte IP datagram.)

In general, the larger the MSS the better, until fragmentation occurs. (This may not always be true. See
Figures 24.3 and 24.4 for a counterexample.) A larger segment size allows more data to be sent in each
segment, amortizing the cost of the IP and TCP headers. When TCP sends a SYN segment, either
because a local application wants to initiate a connection, or when a connection request is received
from another host, it can send an MSS value up to the outgoing interface’s MTU, minus the size of the
fixed TCP and IP headers. For an Ethernet this implies an MSS of up to 1460 bytes. Using IEEE 802.3
encapsulation (Section 2.2), the MSS could go up to 1452 bytes.

The values of 1024 that we’ve seen in this chapter, for connections involving BSD/386 and SVR4, are
because many BSD implementations require the MSS to be a multiple of 512. Other systems, such as
SunOS 4.1.3, Solaris 2.2, and AIX 3.2.2, all announce an MSS of 1460 when both ends are on a local
Ethernet. Measurements in [Mogul 1993] show how an MSS of 1460 provides better performance on
an Ethernet than an MSS of 1024.

If the destination IP address is "nonlocal," the MSS normally defaults to 536. While it’s easy to say
that a destination whose IP address has the same network ID and the same subnet ID as ours is local,
and a destination whose IP address has a totally different network ID from ours is nonlocal, a
destination with the same network ID but a different subnet ID could be either local or nonlocal. Most
implementations provide a configuration option (Appendix E and Figure E.1) that lets the system
administrator specify whether different subnets are local or nonlocal. The setting of this option
determines whether the announced MSS is as large as possible (up to the outgoing interface’s MTU) or

231

the default of 536.

The MSS lets a host limit the size of datagrams that the other end sends it. When combined with the
fact that a host can also limit the size of the datagrams that it sends, this lets a host avoid fragmentation
when the host is connected to a network with a small MTU.

Consider our host slip, which has a SLIP link with an MTU of 296 to the router bsdi. Figure 18.8
shows these systems and the host sun.

[IMAGE]
Figure 18.8 TCP connection from sun to slip showing MSS values.

We initiate a TCP connection from sun to slip and watch the segments using tcpdump. Figure
18.9 shows only the connection establishment (with the window size advertisements removed).

sun.1093 > slip.discard: S 517312000:517312000
1 0.0 (0)
<mss 1460>
slip.discard > sun.1093: S 509556225:509556225
0.10
2 (0)
(0.00)
ack 517312001 <mss 256>
0.10
3 sun.1093 > slip.discard: . ack 1
(0.00)

Figure 18.9 tcpdump output for connection establishment from sun to slip.

The important fact here is that sun cannot send a segment with more than 256 bytes of data, since it
received an MSS option of 256 (line 2). Furthermore, since slip knows that the outgoing interface’s
MTU is 296, even though sun announced an MSS of 1460, it will never send more than 256 bytes of
data, to avoid fragmentation. It’s OK for a system to send less than the MSS announced by the other
end.

This avoidance of fragmentation works only if either host is directly connected to a network with an
MTU of less than 576. If both hosts are connected to Ethernets, and both announce an MSS of 536, but
an intermediate network has an MTU of 296, fragmentation will occur. The only way around this is to
use the path MTU discovery mechanism (Section 24.2).

18.5 TCP Half-Close
TCP provides the ability for one end of a connection to terminate its output, while still receiving data
from the other end. This is called a half-close. Few applications take advantage of this capability, as
we mentioned earlier.

To use this feature the programming interface must provide a way for the application to say "I am
done sending data, so send an end-of-file (FIN) to the other end, but I still want to receive data from
the other end, until it sends me an end-of-file (FIN)."

232

The sockets API supports the half-close, if the application calls shutdown with a second argument of 1, instead of calling
close. Most applications, however, terminate both directions of the connection by calling close.

Figure 18.10 shows a typical scenario for a half-close. We show the client on the left side initiating the
half-close, but either end can do this. The first two segments are the same: a FIN by the initiator,
followed by an ACK of the FIN by the recipient. But it then changes from Figure 18.4, because the
side that receives the half-close can still send data. We show only one data segment, followed by an
ACK, but any number of data segments can be sent. (We talk more about the exchange of data
segments and acknowledgments in Chapter 19.) When the end that received the half-close is done
sending data, it closes its end of the connection, causing a FIN to be sent, and this delivers an
end-of-file to the application that initiated the half-close. When this second FIN is acknowledged, the
connection is completely closed.

[IMAGE]
Figure 18.10 Example of TCP’s half-close.

Why is there a half-close? One example is the Unix rsh(l) command, which executes a command on
another system. The command

sun % rsh bsdi sort < datafile

executes the sort command on the host bsdi with standard input for the rsh command being read
from the file named datafile. A TCP connection is created by rsh between itself and the program
being executed on the other host. The operation of rsh is then simple: it copies standard input
(datafile) to the connection, and copies from the connection to standard output (our terminal).
Figure 18.11 shows the setup. (Remember that a TCP connection is full-duplex.)

[IMAGE]
Figure 18.11 The command: rsh bsdi sort < datafile.

On the remote host bsdi the rshd server executes the sort program so that its standard input and
standard output are both the TCP connection. Chapter 14 of [Stevens 1990] details the Unix process
structure involved, but what concerns us here is the use of the TCP connection and the required use of
TCP’s half-close.

The sort program cannot generate any output until all of its input has been read. All the initial data
across the connection is from the rsh client to the sort server, sending the file to be sorted. When
the end-of-file is reached on the input (datafile), the rsh client performs a half-close on the TCP
connection. The sort server then receives an end-of-file on its standard input (the TCP connection),
sorts the file, and writes the result to its standard output (the TCP connection). The rsh client
continues reading its end of the TCP connection, copying the sorted file to its standard output.

Without a half-close, some other technique is needed to let the client tell the server that the client is
finished sending data, but still let the client receive data from the server. Two connections could be
used as an alternative, but a single connection with a half-close is better.

18.6 TCP State Transition Diagram
We’ve described numerous rules regarding the initiation and termination of a TCP connection. These
rules can be summarized in a state transition diagram, which we show in Figure 18.12.

233

The first thing to note in this diagram is that a subset of the state transitions is "typical." We’ve marked
the normal client transitions with a darker solid arrow, and the normal server transitions with a darker
dashed arrow.

Next, the two transitions leading to the ESTABLISHED state correspond to opening a connection, and
the two transitions leading from the ESTABLISHED state are for the termination of a connection. The
ESTABLISHED state is where data transfer can occur between the two ends in both directions. Later
chapters describe what happens in this state.

We’ve collected the four boxes in the lower left of this diagram within a dashed box and labeled it
"active close." Two other boxes (CLOSE_WAIT and LAST_ACK) are collected in a dashed box with
the label "passive close."

The names of the 11 states (CLOSED, LISTEN, SYN_SENT, etc.) in this figure were purposely
chosen to be identical to the states output by the netstat command. The netstat names, in turn,
are almost identical to the names originally described in RFC 793. The state CLOSED is not really a
state, but is the imaginary starting point and ending point for the diagram.

The state transition from LISTEN to SYN_SENT is legal but is not supported in Berkeley-derived
implementations.

The transition from SYN_RCVD back to LISTEN is valid only if the SYN_RCVD state was entered
from the LISTEN state (the normal scenario), not from the SYN_SENT state (a simultaneous open).
This means if we perform a passive open (enter LISTEN), receive a SYN, send a SYN with an ACK
(enter SYN_RCVD), and then receive a reset instead of an ACK, the end point returns to the LISTEN
state and waits for another connection request to arrive.

[IMAGE]
Figure 18.12 TCP state transition diagram.

Figure 18.13 shows the normal TCP connection establishment and termination, detailing the different
states through which the client and server pass. It is a redo of Figure 18.3 showing only the states.

[IMAGE]
Figure 18.13 TCP states corresponding to normal connection establishment and termination.

We assume in Figure 18.13 that the client on the left side does an active open, and the server on the
right side does a passive open. Although we show the client doing the active close, as we mentioned
earlier, either side can do the active close.

You should follow through the state changes in Figure 18.13 using the state transition diagram in
Figure 18.12, making certain you understand why each state change takes place.

2MSL Wait State
The TIME_WAIT state is also called the 2MSL wait state. Every implementation must choose a value
for the maximum segment lifetime (MSL). It is the maximum amount of time any segment can exist in
the network before being discarded. We know this time limit is bounded, since TCP segments are
transmitted as IP datagrams, and the IP datagram has the TTL field that limits its lifetime.

234

RFC 793 [Postel 1981c] specifies the MSL as 2 minutes. Common implementation values, however, are 30 seconds, 1
minute, or 2 minutes.

Recall from Chapter 8 that the real-world limit on the lifetime of the IP datagram is based on the
number of hops, not a timer.

Given the MSL value for an implementation, the rule is: when TCP performs an active close, and
sends the final ACK, that connection must stay in the TIME_WAIT state for twice the MSL. This lets
TCP resend the final ACK in case this ACK is lost (in which case the other end will time out and
retransmit its final FIN).

Another effect of this 2MSL wait is that while the TCP connection is in the 2MSL wait, the socket pair
defining that connection (client IP address, client port number, server IP address, and server port
number) cannot be reused. That connection can only be reused when the 2MSL wait is over.

Unfortunately most implementations (i.e., the Berkeley-derived ones) impose a more stringent
constraint. By default a local port number cannot be reused while that port number is the local port
number of a socket pair that is in the 2MSL wait. We’ll see examples of this common constraint
below.

Some implementations and APIs provide a way to bypass this restriction. With the sockets API, the SO_REUSEADDR
socket option can be specified. It lets the caller assign itself a local port number that’s in the 2MSL wait, but we’ll see that
the rules of TCP still prevent this port number from being part of a connection that is in the 2MSL wait.

Any delayed segments that arrive for a connection while it is in the 2MSL wait are discarded. Since
the connection defined by the socket pair in the 2MSL wait cannot be reused during this time period,
when we do establish a valid connection we know that delayed segments from an earlier incarnation of
this connection cannot be misinterpreted as being part of the new connection. (A connection is defined
by a socket pair. New instances of a connection are called incarnations of that connection.)

As we said with Figure 18.13, it is normally the client that does the active close and enters the
TIME_WAIT state. The server usually does the passive close, and does not go through the
TIME_WAIT state. The implication is that if we terminate a client, and restart the same client
immediately, that new client cannot reuse the same local port number. This isn’t a problem, since
clients normally use ephemeral ports, and don’t care what the local ephemeral port number is.

With servers, however, this changes, since servers use well-known ports. If we terminate a server that
has a connection established, and immediately try to restart the server, the server cannot assign its
well-known port number to its end point, since that port number is part of a connection that is in a
2MSL wait. It may take from 1 to 4 minutes before the server can be restarted.

We can see this scenario using our sock program. We start the server, connect to it from a client, and
then terminate the server:

235

sun % sock -v -s 6666 start as server, listening on port 6666
(execute client on bsdi that connects to this port)
connection on 140.252.13.33.6666 from 140.252.13.35.1081
^? then type interrupt key to terminate server
sun % sock -s 6666 and immediately try to restart server on same port
sun % netstat let’s check the state of the connection
Active Internet
connections
Local Foreign
Proto Recv-Q Send-Q (state)
Address Address
tcp 0 0 sun.6666 bsdi.1081 TIME_WAIT
many more lines that are deleted

When we try to restart the server, the program outputs an error message indicating it cannot bind its
well-known port number, because it’s already in use (i.e., it’s in a 2MSL wait).

We then immediately execute netstat to see the state of the connection, and verify that it is indeed
in the TIME_WAIT state.

If we continually try to restart the server, and measure the time until it succeeds, we can measure the
2MSL value. On SunOS 4.1.3, SVR4, BSD/386, and AIX 3.2.2, it takes 1 minute to restart the server,
meaning the MSL is 30 seconds. Under Solaris 2.2 it takes 4 minutes to restart the server, implying an
MSL of 2 minutes.

We can see the same error from a client, if the client tries to allocate a port that is part of a connection
in the 2MSL wait (something clients normally don’t do):

sun % sock -v bsdi echo start as client, connect to echo server
connected on 140.252.13.33.1162 to 140.252.13.35.7
hello there type this line
hello there and it’s echoed by the server
^D type end-of-file character to terminate client
sun % sock -b1162 bsdi echo

The first time we execute the client we specify the -v option to see what the local port number is
(1162). The second time we execute the client we specify the -b option, telling the client to assign
itself 1162 as its local port number. As we expect, the client can’t do this, since that port number is
part of a connection that is in a 2MSL wait.

236

We need to reemphasize one effect of the 2MSL wait because we’ll encounter it in Chapter 27 with
FTP, the File Transfer Protocol. As we said earlier, it is a socket pair (that is, the 4-tuple consisting of
a local IP address, local port, remote IP address and remote port) that remains in the 2MSL wait.
Although many implementations allow a process to reuse a port number that is part of a connection
that is in the 2MSL wait (normally with an option named SO_REUSEADDR), TCP cannot allow a new
connection to be created with the same socket pair. We can see this with the following experiment:

start as server, listening on port 6666
sun % sock -v -s 6666
(execute client on bsdi that connects to this port)
connection on 140.252.13.33.6666 from 140.252.13.35.1098
^? then type interrupt key to terminate server
sun % sock -b6666 bsdi 1098 try to start as client with local port 6666
sun % sock -A -b6666 bsdi 1098 try again, this time with -A option
active open error: Address already in use

The first time we run our sock program, we run it as a server on port 6666 and connect to it from a
client on the host bsdi. The client’s ephemeral port number is 1098. We terminate the server so it
does the active close. This causes the 4-tuple of 140.252.13.33 (local IP address), 6666 (local port
number), 140.252.13.35 (foreign IP address), and 1098 (foreign port number) to enter the 2MSL wait
on the server host.

The second time we run the program, we run it as a client and try to specify the local port number as
6666 and connect to host bsdi on port 1098. But the program gets an error when it tries to assign
itself the local port number of 6666, because that port number is part of the 4-tuple that is in the 2MSL
wait state.

To try and get around this error we run the program again, specifying the -A option, which enables the
SO_REUSEADDR option that we mentioned. This lets the program assign itself the port number 6666,
but we then get an error when it tries to issue the active open. Even though it can assign itself the port
number 6666, it cannot create a connection to port 1098 on the host bsdi, because the socket pair
defining that connection is in the 2MSL wait state.

What if we try to establish the connection from the other host? First we must restart the server on sun
with the -A flag, since the local port it needs (6666) is part of a connection that is in the 2MSL wait:

sun % sock -A -s 6666 start as server, listening on port 6666

Then, before the 2MSL wait is over on sun, we start the client on bsdi:

bsdi % sock -bl098 sun 6666
connected on 140.252.13.35.1098 to 140.252.13.33.6666

Unfortunately it works! This is a violation of the TCP specification, but is supported by most
Berkeley-derived implementations. These implementations allow a new connection request to arrive
for a connection that is in the TIME_WAIT state, if the new sequence number is greater than the final

237

sequence number from the previous incarnation of this connection. In this case the ISN for the new
incarnation is set to the final sequence number from the previous incarnation plus 128,000. The
appendix of RFC 1185 [Jacobson, Braden, and Zhang 1990] shows the pitfalls still possible with this
technique.

This implementation feature lets a client and server continually reuse the same port number at each
end for successive incarnations of the same connection, but only if the server does the active close.
We’ll see another example of this 2MSL wait condition in Figure 27.8, with FTP. See Exercise 18.5
also.

Quiet Time Concept
The 2MSL wait provides protection against delayed segments from an earlier incarnation of a
connection from being interpreted as part of a new connection that uses the same local and foreign IP
addresses and port numbers. But this works only if a host with connections in the 2MSL wait does not
crash.

What if a host with ports in the 2MSL wait crashes, reboots within MSL seconds, and immediately
establishes new connections using the same local and foreign IP addresses and port numbers
corresponding to the local ports that were in the 2MSL wait before the crash? In this scenario, delayed
segments from the connections that existed before the crash can be misinterpreted as belonging to the
new connections created after the reboot. This can happen regardless of how the initial sequence
number is chosen after the reboot.

To protect against this scenario, RFC 793 states that TCP should not create any connections for MSL
seconds after rebooting. This is called the quiet time.

Few implementations abide by this since most hosts take longer than MSL seconds to reboot after a crash.

FIN WAIT 2 State
In the FIN_WAIT_2 state we have sent our FIN and the other end has acknowledged it. Unless we
have done a half-close, we are waiting for the application on the other end to recognize that it has
received an end-of-file notification and close its end of the connection, which sends us a FIN. Only
when the process at the other end does this close will our end move from the FIN_WAIT_2 to the
TIME_WAIT state.

This means our end of the connection can remain in this state forever. The other end is still in the
CLOSE_WAIT state, and can remain there forever, until the application decides to issue its close.

Many Berkeley-derived implementations prevent this infinite wait in the FIN_WAIT_2 state as follows. If the application
that does the active close does a complete close, not a half-close indicating that it expects to receive data, then a timer is
set. If the connection is idle for 10 minutes plus 75 seconds, TCP moves the connection into the CLOSED state. A
comment in the code acknowledges that this implementation feature violates the protocol specification.

18.7 Reset Segments
We’ve mentioned a bit in the TCP header named RST for "reset." In general, a reset is sent by TCP
whenever a segment arrives that doesn’t appear correct for the referenced connection. (We use the
term "referenced connection" to mean the connection specified by the destination IP address and port
number, and the source IP address and port number. This is what RFC 793 calls a socket.)

238

Connection Request to Nonexistent Port
A common case for generating a reset is when a connection request arrives and no process is listening
on the destination port. In the case of UDP, we saw in Section 6.5 that an ICMP port unreachable was
generated when a datagram arrived for a destination port that was not in use. TCP uses a reset instead.

This example is trivial to generate-we use the Telnet client and specify a port number that’s not in use
on the destination:

bsdi % telnet: svr4 20000 port 20000 should not be in use
Trying 140.252.13.34...
telnet: Unable to connect to remote host: Connection refused

This error message is output by the Telnet client immediately. Figure 18.14 shows the packet
exchange corresponding to this command.

bsdi.1087 > svr4.20000: S
1 0.0 297416193:297416193(0)
win 4096 <mss 1024> [tos 0x10]
0.003771 svr4.20000 > bsdi.1087: R 0:0(0)
2
(0.0038) ack 297416194 win 0

Figure 18.14 Reset generated by attempt to open connection to nonexistent port.

The values we need to examine in this figure are the sequence number field and acknowledgment
number field in the reset. Because the ACK bit was not on in the arriving segment, the sequence
number of the reset is set to 0 and the acknowledgment number is set to the incoming ISN plus the
number of data bytes in the segment. Although there is no real data in the arriving segment, the SYN
bit logically occupies 1 byte of sequence number space; therefore, in this example the
acknowledgment number in the reset is set to the ISN, plus the data length (0), plus one for the SYN
bit.

Aborting a Connection
We saw in Section 18.2 that the normal way to terminate a connection is for one side to send a FIN.
This is sometimes called an orderly release since the FIN is sent after all previously queued data has
been sent, and there is normally no loss of data. But it’s also possible to abort a connection by sending
a reset instead of a FIN. This is sometimes called an abortive release.

Aborting a connection provides two features to the application: (1) any queued data is thrown away
and the reset is sent immediately, and (2) the receiver of the RST can tell that the other end did an
abort instead of a normal close. The API being used by the application must provide a way to generate
the abort instead of a normal close.

We can watch this abort sequence happen using our sock program. The sockets API provides this
capability by using the "linger on close" socket option (SO_LINGER). We specify the -L option with
a linger time of 0. This causes the abort to be sent when the connection is closed, instead of the normal
FIN. We’ll connect to a server version of our sock program on svr4 and type one line of input:

239

bsdi % sock -LO svr4 8888 this is the client; server shown later
hello, world type one line of input that’s sent to other end
^D type end-of-file character to terminate client

Figure 18.15 shows the tcpdump output for this example. (We have deleted all the window
advertisements in this figure, since they add nothing to the discussion.)

bsdi.1099 > svr4.8888: S
1 0.0 671112193:671112193(0)
<mss 1024>
svr4.8888 > bsdi.1099; S
0.004975
2 3224959489:3224959489(0)
(0.0050)
ack 671112194 <mss 1024>
0.006656
3 bsdi.1099 > svr4.8888: . ack 1
(0.0017)
4.833073
4 bsdi.1099 > svr4.8888: P 1:14(13) ack 1
(4.8264)
5.026224
5 svr4.8888 > bsdi.1099: . ack 14
(0.1932)
9.527634
6 bsdi.1099 > svr4.8888: R 14:14(0) ack 1
(4.5014)

Figure 18.15 Aborting a connection with a reset (RST) instead of a FIN.

Lines 1-3 show the normal connection establishment. Line 4 sends the data line that we typed (12
characters plus the Unix newline character), and line 5 is the acknowledgment of the received data.

Line 6 corresponds to our typing the end-of-file character (Control-D) to terminate the client. Since we
specified an abort instead of a normal close (the -L0 command-line option), the TCP on bsdi sends
an RST instead of the normal FIN. The RST segment contains a sequence number and
acknowledgment number. Also notice that the RST segment elicits no response from the other end-it is
not acknowledged at all. The receiver of the reset aborts the connection and advises the application
that the connection was reset. We get the following error on the server for this exchange:

svr4 % sock -s 8888 run as server, listen on-port 8888
hello, world this is what the client sent over
read error: Connection reset by peer

This server reads from the network and copies whatever it receives to standard output. It normally
ends by receiving an end-of-file notification from its TCP, but here we see that it receives an error
when the RST arrives. The error is what we expect: the connection was reset by the peer.

240

Detecting Half-Open Connections
A TCP connection is said to be half-open if one end has closed or aborted the connection without the
knowledge of the other end. This can happen any time one of the two hosts crashes. As long as there is
no attempt to transfer data across a half-open connection, the end that’s still up won’t detect that the
other end has crashed.

Another common cause of a half-open connection is when a client host is powered off, instead of
terminating the client application and then shutting down the client host. This happens when PCs are
being used to run Telnet clients, for example, and the users power off the PC at the end of the day. If
there was no data transfer going on when the PC was powered off, the server will never la-row that the
client disappeared. When the user comes in the next morning, powers on the PC, and starts a new
Telnet client, a new occurrence of the server is started on the server host. This can lead to many
half-open TCP connections on the server host. (In Chapter 23 we’ll see a way for one end of a TCP
connection to discover that the other end has disappeared using TCP’s keepalive option.)

We can easily create a half-open connection. We’ll execute the Telnet client on bsdi, connecting to
the discard server on svr4. We type one line of input, and watch it go across with tcpdump, and
then disconnect the Ethernet cable on the server’s host, and reboot the server host. This simulates the
server host crashing. (We disconnect the Ethernet cable before rebooting the server to prevent it from
sending a FIN out the open connections, which some TCPs do when they are shut down.) After the
server has rebooted, we reconnect the cable, and try to send another line from the client to the server.
Since the server’s TCP has rebooted, and lost all memory of the connections that existed before it was
rebooted, it knows nothing about the connection that the data segment references. The rule of TCP is
that the receiver responds with a reset.

bsdi % telnet svr4 discard start the client
Trying 140.252.13.34...
Connected to svr4.
Escape character is ’^]’
hi there this line is sent OK.
here is where we reboot the server host
another line and this one elicits a reset
Connection closed by foreign host.

Figure 18.16 shows the tcpdump output for this example. (We have removed from this output the
window advertisements, the type-of-service information, and the MSS announcements, since they add
nothing to the discussion.)

241

bsdi.1102 > svr4.discard:
1 0.0
S1591752193:1591752193(0)
svr4.discard > bsdi.1102:
2 0.004811 (0.0048) S26368001:26368001(0)
ack 1591752194
3 0.006516 (0.0017) bsdi.1102 > svr4.discard: . ack 1
4 5.167679 (5.1612) bsdi.1102 > svr4.discard: P 1:11(10) ack 1
5 5.201662 (0.0340) svr4.discard > bsdi.1102: . ack 11
194.909929
6 bsdi.1102 > svr4.discard: P 11:25(14) ack 1
(189.7083)
194.914957
7 arp who-has bsdi tell svr4
(0.0050)
194.915678
8 arp reply bsdi is-at 0:0:c0:6f:2d:40
(0.0007)
194.918225 svr4.discard > bsdi.1102:
9
(0.0025) R26368002:26368002 (0)

Figure 18.16 Reset in response to data segment on a half-open connection.

Lines 1-3 are the normal connection establishment. Line 4 sends the line "hi there" to the discard
server, and line 5 is the acknowledgment.

At this point we disconnect the Ethernet cable from svr4, reboot it, and reconnect the cable. This
takes almost 190 seconds. We then type the next line of input to the client ("another line") and when
we type the return key the line is sent to the server (line 6 in Figure 18.16). This elicits a response from
the server, but note that since the server was rebooted, its ARP cache is empty, so an ARP request and
reply are required (lines 7 and 8). Then the reset is sent in line 9. The client receives the reset and
outputs that the connection was terminated by the foreign host. (The final message output by the
Telnet client is not as informative as it could be.)

18.8 Simultaneous Open
It is possible, although improbable, for two applications to both perform an active open to each other
at the same time. Each end must transmit a SYN, and the SYNs must pass each other on the network.
It also requires each end to have a local port number that is well known to the other end. This is called
a simultaneous open.

For example, one application on host A could have a local port of 7777 and perform an active open to
port 8888 on host B. The application on host B would have a local port of 8888 and perform an active
open to port 7777 on host A.

This is not the same as connecting a Telnet client on host A to the Telnet server on host B, at the same
time that a Telnet client on host B is connecting to the Telnet server on host A. In this Telnet scenario,
both Telnet servers perform passive opens, not active opens, and the Telnet clients assign themselves
an ephemeral port number, not a port number that is well known to the other Telnet server.

242

TCP was purposely designed to handle simultaneous opens and the rule is that only one connection
results from this, not two connections. (Other protocol suites, notably the OSI transport layer, create
two connections in this scenario, not one.)

When a simultaneous open occurs the state transitions differ from those shown in Figure 18.13. Both
ends send a SYN at about the same time, entering the SYN_SENT state. When each end receives the
SYN, the state changes to SYN_RCVD (Figure 18.12), and each end resends the SYN and
acknowledges the received SYN. When each end receives the SYN plus the ACK, the state changes to
ESTABLISHED. These state changes are summarized in Figure 18.17.

[IMAGE]
Figure 18.17 Segments exchanged during simultaneous open.

A simultaneous open requires the exchange of four segments, one more than the normal three-way
handshake. Also notice that we don’t call either end a client or a server, because both ends act as client
and server.

An Example
It is possible, though hard, to generate a simultaneous open. The two ends must be started at about the
same time, so that the SYNs cross each other. Having a long round-trip time between the two ends
helps, to let the SYNs cross. To do this we’ll execute one end on our host bsdi, and the other end on
the host vangogh.cs.berkeley.edu. Since there is a dialup SLIP link between them, the
round-trip time should be long enough (a few hundred milliseconds) to let the SYNs cross.

One end (bsdi) assigns itself a local port of 8888 (the -b command-line option) and performs an
active open to port 7777 on the other host:

bsdi % sock -v -b8888 vangogh.cs.berkeley.edu 7777
connected on 140.252.13.35.8888 to 128.32.130.2.7777
TCP_MAXSEG = 512
hello, world we type this line
and hi there this line was typed on other end
connection closed by peer this is output when FIN received

The other end is started at about the same time, assigns itself a local port of 7777, and performs an
active open to port 8888:

243

vangogh % sock -v -b7777 bsdi.tuc.noao.edu 8888
connected on 128.32.130.2.7777 to 140.252.13.35.8888
TCP_MAXSEG = 512
hello, world this is typed on the other end
and hi there we type this line
^D and then type our EOF character

We specify the -v flag to our sock program to verify the IP address and port numbers on each end of
the connection. This flag also prints the MSS used by each end of the connection. We also type in one
line on each end, which is sent to the other end and printed, to verify that both ends are indeed talking
to each other.

Figure 18.18 shows the exchange of segments across the connection. (We have deleted some new TCP
options that appear in the original SYN from vangogh, a 4.4BSD system. We describe these newer
options in Section 18.10.) Notice the two SYNs (lines 1 and 2) followed by the two SYNs with ACKs
(lines 3 and 4). These perform the simultaneous open.

Line 5 shows the input line "hello, world" going from bsdi to vangogh, with the acknowledgment
in line 6. Lines 7 and 8 correspond to the line "and hi there" going in the other direction. Lines 9-12
show the normal connection termination.

Many Berkeley-derived implementations do not support the simultaneous open correctly. On these systems, if you can get
the SYNs to cross, you end up with an infinite exchange of segments, each with a SYN and an ACK, in each direction. The
transition from the SYN_SENT state to the SYN_RCVD state in Figure 18.12 is not always tested in many
implementations.

244

bsdi.8888 >vangogh.7777: S
1 0.0 91904001:91904001(0)
win 4096 <mss 512>
vangogh.7777 > bsdi.8888: S
0.213782
2 1058199041:1058199041(0)
(0.2138)
win 8192 <mss 512>
bsdi.8888 > vangogh.7777: S
0.215399
3 91904001:91904001(0)
(0.0016)
ack 1058199042 win 4096 <mss 512>
vangogh.7777 > bsdi.8888: S
0.340405
4 1058199041:1058199041(0)
(0.1250)
ack 91904002 win 8192 <mss 512>
5.633142 bsdi.8888 > vangogh.7777: P 1:14(13) ack 1
5
(5.2927) win 4096
6.100366
6 vangogh.7777 > bsdi.8888: . ack 14 win 8192
(0.4672)
9.640214 vangogh.7777 > bsdi.8888: P 1:14(13) ack 14
7
(3.5398) win 8192
9.796417
8 bsdi.8888 > vangogh.7777: . ack 14 win 4096
(0.1562)
13.060395 vangogh.7777 > bsdi.8888: F 14:14(0) ack 14
9
(3.2640) win 8192
13.061828
10 bsdi.8888 > vangogh.7777: . ack 15 win 4096
(0.0014)
13.079769 bsdi.8888 > vangogh.7777: F 14:14(0) ack 15
11
(0.0179) win 4096
13.299940
12 vangogh.7777 > bsdi.8888: . ack 15 win 8192
(0.2202)

Figure 18.18 Exchange of segments during simultaneous open.

18.9 Simultaneous Close
We said earlier that one side (often, but not always, the client) performs the active close, causing the
first FIN to be sent. It’s also possible for both sides to perform an active close, and the TCP protocol
allows for this simultaneous close.

In terms of Figure 18.12, both ends go from ESTABLISHED to FIN_WAIT_1 when the application
issues the close. This causes both FINs to be sent, and they probably pass each other somewhere in the
network. When the FIN is received, each end transitions from FIN_WAIT_1 to the CLOSING state,
and each state sends its final ACK. When each end receives the final ACK, the state changes to
TIME_WAIT. Figure 18.19 summarizes these state changes.

245

[IMAGE]
Figure 18.19 Segments exchanged during simultaneous close.

With a simultaneous close the same number of segments are exchanged as in the normal close.

18.10 TCP Options
The TCP header can contain options (Figure 17.2). The only options defined in the original TCP
specification are the end of option list, no operation, and the maximum segment size option. We have
seen the MSS option in almost every SYN segment in our examples.

Newer RFCs, specifically RFC 1323 [Jacobson, Braden, and Borman 1992], define additional TCP
options, most of which are found only in the latest implementations. (We describe these new options
in Chapter 24.) Figure 18.20 shows the format of the current TCP options-those from RFC 793 and
RFC 1323.

[IMAGE]
Figure 18.20 TCP options.

Every option begins with a 1-byte kind that specifies the type of option. The options with a kind of 0
and 1 occupy a single byte. The other options have a len byte that follows the kind byte. The length is
the total length, including the kind and len bytes.

The reason for the no operation (NOP) option is to allow the sender to pad fields to a multiple of 4
bytes. If we initiate a TCP connection from a 4.4BSD system, the following TCP options are output by
tcpdump on the initial SYN segment:

<mss 512,nop,wscale 0,nop,nop,timestamp 146647 0>

The MSS option is set to 512, followed by a NOP, followed by the window scale option. The reason
for the first NOP is to pad the 3-byte window scale option to a 4-byte boundary. Similarly, the IO-byte
timestamp option is preceded by two NOPs, to occupy 12 bytes, placing the two 4-byte timestamps
onto 4-byte boundaries.

Four other options have been proposed, with kinds of 4, 5, 6, and 7 called the selective-ACK and echo options. We don’t
show them in Figure 18.20 because the echo options have been replaced with the timestamp option, and selective ACKs, as
currently defined, are still under discussion and were not included in RFC 1323. Also, the T/TCP proposal for TCP
transactions (Section 24.7) specifies three options with kinds of 11, 12, and 13.

18.11 TCP Server Design
We said in Section 1.8 that most TCP servers are concurrent. When a new connection request arrives
at a server, the server accepts the connection and invokes a new process to handle the new client.
Depending on the operating system, various techniques are used to invoke the new server. Under Unix
the common technique is to create a new process using the fork function. Lightweight processes
(threads) can also be used, if supported.

What we’re interested in is the interaction of TCP with concurrent servers. We need to answer the
following questions: how are the port numbers handled when a server accepts a new connection
request from a client, and what happens if multiple connection requests arrive at about the same time?

246

TCP Server Port Numbers
We can see how TCP handles the port numbers by watching any TCP server. We’ll watch the Telnet
server using the netstat command. The following output is on a system with no active Telnet
connections. (We have deleted all the lines except the one showing the Telnet server.)

sun % netstat -a -n -f inet
Active Internet connections (including servers)
tcp 0 0 *.23 *.* LISTEN

The -a flag reports on all network end points, not just those that are ESTABLISHED. The -n flag
prints IP addresses as dotted-decimal numbers, instead of trying to use the DNS to convert the address
to a name, and prints numeric port numbers (e.g., 23) instead of service names (e.g., Telnet). The -f
inet option reports only TCP and UDP end points.

The local address is output as *.23, where the asterisk is normally called the wildcard character. This
means that an incoming connection request (i.e., a SYN) will be accepted on any local interface. If the
host were multihomed, we could specify a single IP address for the local IP address (one of the host’s
IP addresses), and only connections received on that interface would be accepted. (We’ll see an
example of this later in this section.) The local port is 23, the well-known port number for Telnet.

The foreign address is output as *.*, which means the foreign IP address and foreign port number are
not known yet, because the end point is in the LISTEN state, waiting for a connection to arrive.

We now start a Telnet client on the host slip (140.252.13.65) that connects to this server. Here are
the relevant lines from the netstat output:

tcp 0 0 140.252.13.33.23 140.252.13.65.1029 ESTABLISHED
tcp 0 0 *.23 *.* LISTEN

The first line for port 23 is the ESTABLISHED connection. All four elements of the local and foreign
address are filled in for this connection: the local IP address and port number, and the foreign IP
address and port number. The local IP address corresponds to the interface on which the connection
request arrived (the Ethernet interface, 140.252.13.33).

The end point in the LISTEN state is left alone. This is the end point that the concurrent server uses to
accept future connection requests. It is the TCP module in the kernel that creates the new end point in
the ESTABLISHED state, when the incoming connection request arrives and is accepted. Also notice
that the port number for the ESTABLISHED connection doesn’t change: it’s 23, the same as the
LISTEN end point.

We now initiate another Telnet client from the same client (slip) to this server. Here is the relevant
netstat output:

247

tcp 0 0 140.252.13.33.23 140.252.13.65.1030 ESTABLISHED
tcp 0 0 140.252.13.33.23 140.252.13.65.1029 ESTABLISHED
tcp 0 0 *.23 *.* LISTEN

We now have two ESTABLISHED connections from the same host to the same server. Both have a
local port number of 23. This is not a problem for TCP since the foreign port numbers are different.
They must be different because each of the Telnet clients uses an ephemeral port, and the definition of
an ephemeral port is one that is not currently in use on that host (slip).

This example reiterates that TCP demultiplexes incoming segments using all four values that comprise
the local and foreign addresses: destination IP address, destination port number, source IP address, and
source port number. TCP cannot determine which process gets an incoming segment by looking at the
destination port number only. Also, the only one of the three end points at port 23 that will receive
incoming connection requests is the one in the LISTEN state. The end points in the ESTABLISHED
state cannot receive SYN segments, and the end point in the LISTEN state cannot receive data
segments.

Next we initiate a third Telnet client, from the host solaris that is across the SLIP link from sun,
and not on its Ethernet.

tcp 0 0 140.252.1.29.23 140.252.1.32.34603 ESTABLISHED
tcp 0 0 140.252.13.33.23 140.252.13.65.1030 ESTABLISHED
tcp 0 0 140.252.13.33.23 140.252.13.65.1029 ESTABLISHED
tcp 0 0 *.23 *.* LISTEN

The local IP address of the first ESTABLISHED connection now corresponds to the interface address
of SLIP link on the multihomed host sun (140.252.1.29).

Restricting Local IP Address
We can see what happens when the server does not wildcard its local IP address, setting it to one
particular local interface address instead. If we specify an IP address (or host-name) to our sock
program when we invoke it as a server, that IP address becomes the local IP address of the listening
end point. For example

sun % sock -s 140.252.1.29 8888

restricts this server to connections arriving on the SLIP interface (140.252.1.29). The netstat
output reflects this:

248

Foreign
Address
tcp 0 0 140.252.1.29.8888 *.* LISTEN

If we connect to this server across the SLIP link, from the host solaris, it works.

tcp 0 0 140.252.1.29.8888 140.252.1.32.34614 ESTABLISHED
tcp 0 0 140.252.1.29.8888 *.* LISTEN

But if we try to connect to this server from a host on the Ethernet (140.252.13), the connection request
is not accepted by the TCP module. If we watch it with tcpdump the SYN is responded to with an
RST, as we show in Figure 18.21.

bsdi.l026 > sun.8888: S
1 0.0 3657920001:3657920001(0)
win 4096 <mss 1024>
0.000859 sun.8888 > bsdi.l026: R 0:0(0) ack 3657920002
2
(0.0009) win 0

Figure 18.21 Rejection of a connection request based on local IP address of server.

The server application never sees the connection request - the rejection is done by the kernel’s TCP
module, based on the local IP address specified by the application.

Restricting Foreign IP Address
In Section 11.12 we saw that a UDP server can normally specify the foreign IP address and foreign
port, in addition to specifying the local IP address and local port. The interface functions shown in
RFC 793 allow a server doing a passive open to have either a fully specified foreign socket (to wait for
a particular client to issue an active open) or a unspecified foreign socket (to wait for any client).

Unfortunately, most APIs don’t provide a way to do this. The server must leave the foreign socket
unspecified, wait for the connection to arrive, and then examine the IP address and port number of the
client.

Figure 18.22 summarizes the three types of address bindings that a TCP server can establish for itself.
In all cases, lport is the server’s well-known port and localIP must be the IP address of a local
interface. The ordering of the three rows in the table is the order that the TCP module applies when
trying to determine which local end point receives an incoming connection request. The most specific
binding (the first row, if supported) is tried first, and the least specific (the last row with both IP
addresses wild-carded) is tried last.

249

Local Address Foreign Address Description
localIP.lport foreignIP.fport restricted to one client (normally not supported)
localIP.lport *.* restricted to connections arriving on one local interface: localIP
*.lport ** receives all connections sent to lport

Figure 18.22 Specification of local and foreign IP addresses and port number for TCP server.

Incoming Connection Request Queue
A concurrent server invokes a new process to handle each client, so the listening server should always
be ready to handle the next incoming connection request. That’s the underlying reason for using
concurrent servers. But there is still a chance that multiple connection requests arrive while the
listening server is creating a new process, or while the operating system is busy running other higher
priority processes. How does TCP handle these incoming connection requests while the listening
application is busy? In Berkeley-derived implementations the following rules apply.

1. Each listening end point has a fixed length queue of connections that have been accepted by TCP
(i.e., the three-way handshake is complete), but not yet accepted by the application.

Be careful to differentiate between TCP accepting a connection and placing it on this queue, and
the application taking the accepted connection off this queue.

2. The application specifies a limit to this queue, commonly called the backlog. This backlog must
be between 0 and 5, inclusive. (Most applications specify the maximum value of 5.)

3. When a connection request arrives (i.e., the SYN segment), an algorithm is applied by TCP to the
current number of connections already queued for this listening end point, to see whether to
accept the connection or not. We would expect the backlog value specified by the application to
be the maximum number of queued connections allowed for this end point, but it’s not that
simple. Figure 18.23 shows the relationship between the backlog value and the real maximum
number of queued connections allowed by traditional Berkeley systems and Solaris 2.2.

Backlog value Max # of queued connections
Traditional BSD Solaris 2.2
0 1 0
1 2 1
2 4 2
3 5 3
4 7 4
5 8 5

Figure 18.23 Maximum number of accepted connections allowed for listening end point.

Keep in mind that this backlog value specifies only the maximum number of queued connections
for one listening end point, all of which have already been accepted by TCP and are waiting to be
accepted by the application. This backlog has no effect whatsoever on the maximum number of
established connections allowed by the system, or on the number of clients that a concurrent
server can handle concurrently.

250

The Solaris values in this figure are what we expect. The traditional BSD values are (for some unknown reason) the
backlog value times 3, divided by 2, plus 1.

4. If there is room on this listening end point’s queue for this new connection (based on Figure
18.23), the TCP module ACKs the SYN and completes the connection. The server application with the
listening end point won’t see this new connection until the third segment of the three-way handshake
is received. Also, the client may think the server is ready to receive data when the client’s active open
completes successfully, before the server application has been notified of the new connection. (If this
happens, the server’s TCP just queues the incoming data.)

5. If there is not room on the queue for the new connection, TCP just ignores the received SYN.
Nothing is sent back (i.e., no RST segment). If the listening server doesn’t get around to accepting
some of the already accepted connections that have filled its queue to the limit, the client’s active open
will eventually time out.

We can see this scenario take place with our sock program. We invoke it with a new option (-O) that
tells it to pause after creating the listening end point, before accepting any connection requests. If we
then invoke multiple clients during this pause period, it should cause the server’s queue of accepted
connections to fill, and we can see what happens with tcpdump.

bsdi % sock -a -v -q1 -O30 7777

The -q1 option sets the backlog of the listening end point to 1, which for this traditional BSD system
should allow two pending connection requests (Figure 18.23). The -O30 option causes the program to
sleep for 30 seconds before accepting any client connections. This gives us 30 seconds to start some
clients, to fill the queue. We’ll start four clients on the host sun.

Figure 18.24 shows the tcpdump output, starting with the first SYN from the first client. (We have
removed the window size advertisements and MSS announcements. We have also marked the client
port numbers in bold when the TCP connection is established-the three-way handshake.)

The first client’s connection request from port 1090 is accepted by TCP (segments 1-3). The second
client’s connection request from port 1091 is also accepted by TCP (segments 4-6). The server
application is still asleep, and has not accepted either connection yet. Everything has been done by the
TCP module in the kernel. Also, the two clients have returned successfully from their active opens,
since the three-way handshakes are complete.

251

sun.1090 > bsdi.7777: S
1 0.0
1617152000:1617152000(0)
bsdi.7777 > sun.1090: S
0.002310 (
2 4164096001:4164096001(0)
0.0023)
ack 1617152001
0.003098 (
3 sun.1090 > bsdi.7777: . ack 1
0.0008)
4.291007 ( sun.1091 > bsdi.7777: S
4
4.2879) 1617792000:1617792000(0)
4.293349 (
5 S 4164672001:4164672001(0) ack 1617792001
0.0023)
4.294167 (
6 sun.1091 > bsdi.7777: . ack 1
0.0008)
7.131981 ( sun.1092 > bsdi.7777: S
7
2.8378) 1618176000:1618176000 (0)
10.556787 ( sun.1093 > bsdi.7777: S
8
3..4248) 1618688000:1618688000 (0)
12.695916 ( sun.1092 > bsdi.7777: S
9
2..1391) 1618176000:1618176000 (0)
16.195772 ( sun.1093 > bsdi.7777: S
10
3..4999) 1618688000:1618688000 (0)
24.695571 ( sun.1092 > bsdi.7777: S
11
8..4998) 1618176000:1618176000 (0)
28.195454 ( sun.1093 > bsdi.7777: S
12
3.4999) 1618688000:1618688000 (0)
bsdi.7777 > sun.1093: S
28.197810 (
13 4167808001:4167808001 (0)
0.0024)
ack 1618688001
28.198639 (
14 sun.1093 > bsdi.7777: ack 1
0.0008)
48.694931 sun.1092 > bsdi.7777: S
15
(20.4963) 1618176000:1618176000(0)
bsdi.7777 > sun.1092: S 4190496001: 417 049
48.697292 (
16 6001(0)
0.0024)
ack 1618176001
48.698145 (
17 sun.1092 > bsdi.7777: ack 1
0.0009)

252

Figure 18.24 tcpdump output for backlog example.

We try to start a third client in segment 7 (port 1092), and a fourth in segment 8 (port 1093). TCP
ignores both SYNs since the queue for this listening end point is full. Both clients retransmit their
SYNs in segments 9, 10, 11, 12, and 15. The fourth client’s third retransmission is accepted (segments
12-14) because the server’s 30-second pause is over, causing the server to remove the two connections
that were accepted, emptying its queue. (The reason it appears this connection was accepted by the
server at the time 28.19, and not at a time greater than 30, is because it took a few seconds to start the
first client [segment 1, the starting time point in the output] after starting the server.) The third client’s
fourth retransmission is then accepted (segments 15-17). The fourth client connection (port 1093) is
accepted by the server before the third client connection (port 1092) because of the timing interactions
between the server’s 30-second pause and the client’s retransmissions.

We would expect the queue of accepted connections to be passed to the application in FIFO (first-in, first-out) order. That
is, after TCP accepts the connections on ports 1090 and 1091, we expect the application to receive the connection on port
1090 first, and then the connection on port 1091. But a bug has existed for years in many Berkeley-derived
implementations causing them to be returned in a LIFO (last-in, first-out) order instead. Vendors have recently started
fixing this bug, but it still exists in systems such as SunOS 4.1.3.

TCP ignores the incoming SYN when the queue is full, and doesn’t respond with an RST, because this
is a soft error, not a hard error. Normally the queue is full because the application or the operating
system is busy, preventing the application from servicing incoming connections. This condition could
change in a short while. But if the server’s TCP responded with a reset, the client’s active open would
abort (which is what we saw happen if the server wasn’t started). By ignoring the SYN, the server
forces the client TCP to retransmit the SYN later, hoping that the queue will then have room for the
new connection.

A subtle point in this example, which is found in most TCP/IP implementations, is that TCP accepts an
incoming connection request (i.e., a SYN) if there is room on the listener’s queue, without giving the
application a chance to see who it’s from (the source IP address and source port number). This is not
required by TCP, it’s just the common implementation technique (i.e., the way the Berkeley sources
have always done it). If an API such as TLI (Section 1.15) gives the application a way to learn when a
connection request arrives, and then allows the application to choose whether to accept the connection
or not, be aware that with TCP, when the application is supposedly told that the connection has just
arrived, TCP’s three-way handshake is over! Other transport layers may be implemented to provide
this separation to the application between arrival and acceptance (i.e., the OSI transport layer) but not
TCP.

Solaris 2.2 provides an option that prevents TCP from accepting an incoming connection request until the application says
so (tcp_eager_listeners in Section E.4).

This behavior also means that a TCP server has no way to cause a client’s active open to fail. When a
new client connection is passed to the server application, TCP’s three-way handshake is over, and the
client’s active open has completed successfully. If the server then looks at the client’s IP address and
port number, and decides it doesn’t want to service this client, all the server can do is either close the
connection (causing a FIN to be sent) or reset the connection (causing an RST to be sent). In either
case the client thought everything was OK when its active open completed, and may have already sent
a request to the server.

253

18.12 Summary
Before two processes can exchange data using TCP, they must establish a connection between
themselves. When they’re done they terminate the connection. This chapter has provided a detailed
look at how connections are established using a three-way handshake, and terminated using four
segments.

We used tcpdump to show all the fields in the TCP header. We’ve also seen how a connection
establishment can time out, how resets are sent, what happens with a half-open connection, and how
TCP provides a half-close, simultaneous opens, and simultaneous closes.

Fundamental to understanding the operation of TCP is its state transition diagram. We’ve followed
through the steps involved in connection establishment and termination, and the state transitions that
take place. We also looked at the implications of TCP’s connection establishment on the design of
concurrent TCP servers.

A TCP connection is uniquely defined by a 4-tuple: the local IP address, local port number, foreign IP
address, and foreign port number. Whenever a connection is terminated, one end must maintain
knowledge of the connection, and we saw that the TIME_WAIT state handles this. The rule is that the
end that does the active open enters this state for twice the implementation’s MSL.

Exercises
18.1 In Section 18.2 we said that the initial sequence number (ISN) normally starts at 1 and is
incremented by 64,000 every half-second and every time an active open is performed. This would
imply that the low-order three digits of the ISN would always be 001. But in Figure 18.3 these
low-order three digits are 521 in each direction. What’s going on?

18.2 In Figure 18.15 we typed 12 characters and saw 13 bytes sent by TCP. In Figure 18.16 we typed
eight characters but TCP sent 10 bytes. Why was 1 byte added in the first case, but 2 bytes in the
second case?

18.3 What’s the difference between a half-open connection and a half-closed connection?

18.4 If we start our sock program as a server, and then terminate it (without having a client connect to
it), we can immediately restart the server. This implies that it doesn’t go through the 2MSL wait state.
Explain this in terms of the state transition diagram.

18.5 In Section 18.6 we showed that a client cannot reuse the same local port number while that port is
part of a connection in the 2MSL wait. But if we run our sock program twice in a row as a client,
connecting to the daytime server, we can reuse the same local port number. Additionally, we’re able to
create a new incarnation of a connection that should be in the 2MSL wait. What’s going on?

sun % sock -v bsdi daytime
connected on 140.252.13.33.1163 to 140.252.13.35.13
Wed Jul 7 07:54:51 1993
connection closed by peer

sun % sock -v -bll63 bsdi daytime reuse same local port number
connected on 140.252.13.33.1163 to 140.252.13.35.13
Wed Jul 707:55:01 1993
connection closed by peer

254

18.6 At the end of Section 18.6 when describing the FIN_WAIT_2 state, we mentioned that many
implementations move a connection from this state into the CLOSED state if the application did a
complete close (not a half-close) after just over 11 minutes. If the other end (in the CLOSE_WAIT
state) waited 12 minutes before issuing its close (i.e., sending its FIN), what would its TCP get in
response to the FIN?

18.7 Which end of a telephone conversation does the active open, and which does the passive open?
Are simultaneous opens allowed? Are simultaneous closes allowed?

18.8 In Figure 18.6 we don’t see an ARP request or an ARP reply. Obviously the hardware address for
host svr4 must be in the ARP cache on bsdi. What would change in this figure if this ARP cache
entry was not present?

18.9 Explain the following tcpdump output. Compare it with Figure 18.13.

solaris.32990 > bsdi.discard: S
1 0.0 40140288:40140288(0)
win 8760 <mss 1460>
bsdi.discard > solaris.32990: S
0.003295
2 4208081409:4208081409 (0)
(0.0033)
ack 40140289 win 4096 <mss 1024>
0.419991 solaris.32990 > bsdi.discard: P 1:257(256) ack
3
(0.4167) 1 win 9216
0.449852 solaris.32990 > bsdi.discard: F 257:257(0) ack
4
(0.0299) 1 win 9216
0.451965 bsdi.discard > solaris.32990: . ack 258 win
5
(0.0021) 3840
0.464569 bsdi.discard > solaris.32990: F 1:1(0) ack 258
6
(0.0126) win 4096
0.720031
7 solaris.32990 > bsdi.discard: . ack 2 win 9216
(0.2555)

18.10 Why doesn’t the server in Figure 18.4 combine the ACK of the client’s FIN with its own FIN,
reducing the number of segments to three?

18.11 In Figure 18.16 why is the sequence number of the RST 26368002?

18.12 Does TCP’s querying the link layer for the MTU violate the spirit of layering?

18.13 Assume in Figure 14.16 that each DNS query is issued using TCP instead of UDP. How many
packets are exchanged?

18.14 With an MSL of 120 seconds, what is the maximum at which a system can initiate new
connections and then do an active close?

255

18.15 Read RFC 793 to see what happens when an end point that is in the TIME_WAIT state receives
a duplicate of the FIN that placed it into this state.

18.16 Read RFC 793 to see what happens when an end point that is in the TIME_WAIT state receives
an RST.

18.17 Read the Host Requirements RFC to obtain the definition of a half-duplex TCP close.

18.18 In Figure 1.8 we said that incoming TCP segments are demultiplexed based on the destination
TCP port number. Is that correct?

256

TCP Interactive Data Flow
19.1 Introduction
The previous chapter dealt with the establishment and termination of TCP connections. We now
examine the transfer of data using TCP.

Studies of TCP traffic, such as [Caceres et al. 1991], usually find that on a packet-count basis about
half of all TCP segments contain bulk data (FTP, electronic mail, Usenet news) and the other half
contain interactive data (Telnet and Rlogin, for example). On a byte-count basis the ratio is around
90% bulk data and 10% interactive, since bulk data segments tend to be full sized (normally 512 bytes
of user data), while interactive data tends to be much smaller. (The above-mentioned study found that
90% of Telnet and Rlogin packets carry less than 10 bytes of user data.)

TCP obviously handles both types of data, but different algorithms come into play for each. In this
chapter we’ll look at interactive data transfer, using the Rlogin application. We’ll see how delayed
acknowledgments work and how the Nagle algorithm reduces the number of small packets across wide
area networks. The same algorithms apply to Telnet. In the next chapter we’ll look at bulk data
transfer.

19.2 Interactive Input
Let’s look at the flow of data when we type an interactive command on an Rlogin connection. Many
newcomers to TCP/IP are surprised to find that each interactive keystroke normally generates a data
packet. That is, the keystrokes are sent from the client to the server 1 byte at a time (not one line at a
time). Furthermore, Rlogin has the remote system (the server) echo the characters that we (the client)
type. This could generate four segments: (1) the interactive keystroke from the client, (2) an
acknowledgment of the keystroke from the server, (3) the echo of the keystroke from the server, and
(4) an acknowledgment of the echo from the client. Figure 19.1 shows this flow of data.

Figure 19.1 One possible way to do remote echo of interactive keystroke.

Normally, however, segments 2 and 3 are combined-the acknowledgment of the keystroke is sent
along with the echo. We describe the technique that combines these (called delayed acknowledgments)
in the next section.

257

We purposely use Rlogin for the examples in this chapter because it always sends one character at a
time from the client to the server. When we describe Telnet in Chapter 26, we’ll see that it has an
option that allows lines of input to be sent from the client to the server, which reduces the network
load.

Figure 19.2 shows the flow of data when we type the five characters daten. (We do not show the
connection establishment and we have removed all the type-of-service output. BSD/386 sets the TOS
for an Rlogin connection for minimum delay.) Line 1 sends the character d from the client to the
server. Line 2 is the acknowledgment of this character and its echo. (This is combining the middle two
segments in Figure 19.1.) Line 3 is the acknowledgment of the echoed character. Lines 4-6 correspond
to the character a, lines 7-9 to the character t, and lines 10-12 to the character e. The fractional second
delays between lines 3-4, 6-7, 9-10, and 12-13 are the human delays between typing each character.

Notice that lines 13-15 are slightly different. One character is sent from the client to the server (the
Unix newline character, from our typing the RETURN key) but two characters are echoed. These two
characters are a carriage return and linefeed (CR/LF), to move the cursor back to the left and space
down one line.

Line 16 is the output of the date command from the server. The 30 bytes are composed of the
following 28 characters

Sat Feb 6 07:52:17 MST 1993
bsdi.1023 > svr4.login: P 0:1(1) ack 1 win
1 0.0
4096
0.016497 svr4.login > bsdi.1023: P 1:2(1) ack 1 win
2
(0.0165) 4096
0.139955
3 bsdi.1023 > svr4.login: . ack 2 win 4096
(0.1235)
0.458037 bsdi.1023 > svr4.login: P 1:2(1) ack 2 win
4
(0.3181) 4096
5
(0.0163) 4096
0.539943
(0.0656)
7
(0.2746) 4096
8
(0.0165) 4096
0.940112
(0.1090)
10
(0.2512) 4096

258

11
(0.0164) 4096
1.339994
(0.1323)
13
(0.3407) 4096
14
(0.0173) 4096
1.739974
(0.0420)
16
(0.0599) 4096
1.940176
(0.1403)
18
(0.0042) 4096
2.140110
(0.1958)

Figure 19.2 TCP segments when date typed on Rlogin connection.

plus a CR/LF pair at the end. The next 7 bytes sent from the server to the client (line 18) are the
client’s prompt on the server host: svr4 % . Line 19 acknowledges these 7 bytes.

Notice how the TCP acknowledgments operate. Line 1 sends the data byte with the sequence number
0. Line 2 ACKs this by setting the acknowledgment sequence number to 1, the sequence number of the
last successfully received byte plus one. (This is also called the sequence number of the next expected
byte.) Line 2 also sends the data byte with a sequence number of 1 from the server to the client. This is
ACKed by the client in line 3 by setting the acknowledged sequence number to 2.

19.3 Delayed Acknowledgments
There are some subtle points in Figure 19.2 dealing with timing that we’ll cover in this section. Figure
19.3 shows the time line for the exchange in Figure 19.2. (We have deleted all the window
advertisements from this time line, and have added a notation indicating what data is being
transferred.)

We have labeled the seven ACKs sent from bsdi to svr4 as delayed ACKs. Normally TCP does not
send an ACK the instant it receives data. Instead, it delays the ACK, hoping to have data going in the
same direction as the ACK, so the ACK can be sent along with the data. (This is sometimes called
having the ACK piggyback with the data.) Most implementations use a 200-ms delay-that is, TCP will
delay an ACK up to 200 ms to see if there is data to send with the ACK.

259

[IMAGE]
Figure 19.3 Time line of data flow for date command typed on an rlogin connection.

If we look at the time differences between bsdi receiving the data and sending the ACK, they appear
to be random: 123.5, 65.6, 109.0, 132.3,42.0, 140.3, and 195.8 ms. Look instead at the actual times
(starting from 0) when the ACKs are sent: 139.9, 539.9, 940.1, 1339.9, 1739.9, 1940.1, and 2140.1 ms.
(We have marked these with an asterisk to the left of the time in Figure 19.3.) There is a multiple of
200 ms between these times. What is happening here is that TCP has a timer that goes off every 200
ms, but it goes off at fixed points in time-every 200 ms relative to when the kernel was bootstrapped.
Since the data being acknowledged arrives randomly (at times 16.4, 474.3, 831.1, etc.), TCP asks to be
notified the next time the kernel’s 200-ms timer expires. This can be anywhere from 1 to 200 ms in the
future.

If we look at how long it takes svr4 to generate the echo of each character it receives, the times are
16.5, 16.3, 16.5, 16.4, and 17.3 ms. Since this time is less than 200 ms, we never see a delayed ACK
on that side. There is always data ready to be sent before the delayed ACK timer expires. (We could
still see a delayed ACK if the wait period, about 16 ms, crosses one of the kernel’s 200-ms clock tick
boundaries. We just don’t see any of these in this example.)

We saw this same scenario in Figure 18.7 with the 500-ms TCP timer used when detecting a timeout.
Both TCP timers, the 200- and 500-ms timers, go off at times relative to when the kernel was
bootstrapped. Whenever TCP sets a timer, it can go off anywhere between 1-200 or 1-500 ms in the
future.

The Host Requirements RFC states that TCP should implement a delayed ACK but the delay must be
less than 500 ms.

19.4 Nagle Algorithm
We saw in the previous section that 1 byte at a time normally flows from the client to the server across
an Rlogin connection. This generates 41-byte packets: 20 bytes for the IP header, 20 bytes for the TCP
header, and 1 byte of data. These small packets (called tinygrams) are normally not a problem on
LANs, since most LANs are not congested, but these tinygrams can add to congestion on wide area
networks. A simple and elegant solution was proposed in RFC 896 [Nagle 1984], called the Nagle
algorithm.

This algorithm says that a TCP connection can have only one outstanding small segment that has not
yet been acknowledged. No additional small segments can be sent until the acknowledgment is
received. Instead, small amounts of data are collected by TCP and sent in a single segment when the
acknowledgment arrives. The beauty of this algorithm is that it is self-clocking: the faster the ACKs
come back, the faster the data is sent. But on a slow WAN, where it is desired to reduce the number of
tinygrams, fewer segments are sent. (We’ll see in Section 22.3 that the definition of "small" is less
than the segment size.)

We saw in Figure 19.3 that the round-trip time on an Ethernet for a single byte to be sent,
acknowledged, and echoed averaged around 16 ms. To generate data faster than this we would have to
be typing more than 60 characters per second. This means we rarely encounter this algorithm when
sending data between two hosts on a LAN.

260

Things change, however, when the round-trip tune (RTT) increases, typically across a WAN. Let’s
look at an Rlogin connection between our host slip and the host vangogh.cs.berkeley.edu.
To get out of our network (see inside front cover), two SLIP links must be traversed, and then the
Internet is used. We expect much longer round-trip times. Figure 19.4 shows the time line of some
data flow while characters were being typed quickly on the client (similar to a fast typist). (We have
removed the type-of-service information, but have left in the window size advertisements.)

Figure 19.4 Dataflow using rlogin between slip and vangogh.cs.berkeley.edu.

The first thing we notice, comparing Figure 19.4 with Figure 19.3, is the lack of delayed ACKs from
slip to vangogh. This is because there is always data ready to send before the delayed ACK timer
expires.

Next, notice the various amounts of data being sent from the left to the right: 1, 1, 2, 1, 2, 2, 3, 1, and 3
bytes. This is because the client is collecting the data to send, but doesn’t send it until the previously
sent data has been acknowledged. By using the Nagle algorithm only nine segments were used to send
16 bytes, instead of 16 segments.

261

Segments 14 and 15 appear to contradict the Nagle algorithm, but we need to look at the sequence
numbers to see what’s really happening. Segment 14 is in response to the ACK received in segment
12, since the acknowledged sequence number is 54. But before this data segment is sent by the client,
segment 13 arrives from the server. Segment 15 contains the ACK of segment 13, sequence number
56. So the client is obeying the Nagle algorithm, even though we see two back-to-back data segments
from the client to the server.

Also notice in Figure 19.4 that one delayed ACK is present, but it’s from the server to the client
(segment 12). We are assuming this is a delayed ACK since it contains no data. The server must have
been busy at this time, so that the Rlogin server was not able to echo the character before the server’s
delayed ACK timer expired.

Finally, look at the amounts of data and the sequence numbers in the final two segments. The client
sends 3 bytes of data (numbered 18, 19, and 20), then the server acknowledges these 3 bytes (the ACK
of 21 in the final segment) but sends back only 1 byte (numbered 59). What’s happening here is that
the server’s TCP is acknowledging the 3 bytes of data once it has received them correctly, but it won’t
have the echo of these 3 bytes ready to send back until the Rlogin server sends them. This shows that
TCP can acknowledge received data before the application has read and processed that data. The TCP
acknowledgment just means TCP has correctly received the data. We also have an indication that the
server process has not read these 3 bytes of data because the advertised window in the final segment is
8189, not 8192.

Disabling the Nagle Algorithm
There are times when the Nagle algorithm needs to be turned off. The classic example is the X
Window System server (Section 30.5): small messages (mouse movements) must be delivered without
delay to provide real-time feedback for interactive users doing certain operations.

Here we’ll show another example that’s easier to demonstrate-typing one of the terminal’s special
function keys during an interactive login. The function keys normally generate multiple bytes of data,
often beginning with the ASCII escape character. If TCP gets the data 1 byte at a time, it’s possible for
it to send the first byte (the ASCII ESC) and then hold the remaining bytes of the sequence waiting for
the ACK of this byte. But when the server receives this first byte it doesn’t generate an echo until the
remaining bytes are received. This often triggers the delayed ACK algorithm on the server, meaning
that the remaining bytes aren’t sent for up to 200 ms. This can lead to noticeable delays to the
interactive user.

The sockets API uses the TCP_NODELAY socket option to disable the Nagle algorithm.

The Host Requirements RFC states that TCP should implement the Nagle algorithm but there must be
a way for an application to disable it on an individual connection.

An Example
We can see this interaction between the Nagle algorithm and keystrokes that generate multiple bytes.
We establish an Rlogin connection from our host slip to the host vangogh.cs.berkeley.edu.
We then type the Fl function key, which generates 3 bytes: an escape, a left bracket, and an M. We
then type the F2 function key, which generates another 3 bytes. Figure 19.5 shows the tcpdump output.
(We have removed the type-of-service information and the window advertisements.)

262

type Fl key
1 0.0 slip.1023 > vangogh. login: P 1:2(1) ack 2
2 0.250520 (0.2505) vangogh.login > slip.1023: P 2:4(2) ack 2
3 0.251709 (0.0012) slip.1023 > vangogh.login: P 2:4(2) ack 4
5 0.588694 (0.0984) slip.1023 > vangogh.login: . ack 6
type F2 key

Figure 19.5 Watching the Nagle algorithm when typing characters that generate multiple bytes of
data.

Figure 19.6 shows the time line for this exchange. At the bottom of this figure we show the 6 bytes
going from the client to the server with their sequence numbers, and the 8 bytes of echo being
returned.

When the first byte of input is read by the rlogin client and written to TCP, it is sent by itself as
segment 1. This is the first of the 3 bytes generated by the Fl key. Its echo is returned in segment 2,
and only then are the next 2 bytes sent (segment 3). The echo of the second 2 bytes is received in
segment 4 and acknowledged in segment 5.

The reason the echo of the first byte occupies 2 bytes (segment 2) is because the ASCII escape
character is echoed as 2 bytes: a caret and a left bracket. The next 2 bytes of input, a left bracket and
an M, are echoed as themselves.

The same exchange occurs when the next special function key is typed (segments 6-10). As we expect,
the time difference between segments 5 and 10 (slip sending the acknowledgment of the echo) is a
multiple of 200 ms, since both ACKs are delayed.

263

Figure 19.6 Time line for Figure 19.5 (watching the Nagle algorithm).

We now repeat this same example using a version of rlogin that has been modified to turn off the
Nagle algorithm. Figure 19.7 shows the tcpdump output. (Again, we have deleted the type-of-service
information and the window advertisements.)

264

type Fl key
1 0.0 slip.1023 > vangogh.login: P 1:2(1) ack 2
type F2 key

Figure 19.7 Disabling the Nagle algorithm during an Rlogin session.

It is instructive and more enlightening to take this output and construct the time line, knowing that
some of the segments are crossing in the network. Also, this example requires careful examination of
the sequence numbers, to follow the data flow. We show this in Figure 19.8. We have numbered the
segments to correspond with the numbering in the tcpdump output in Figure 19.7.

The first change we notice is that all 3 bytes are sent when they’re ready (segments 1,2, and 3). "There
is no delay-the Nagle algorithm has been disabled.

The next packet we see in the tcpdump output (segment 4) contains byte 5 from the server with an
ACK 4. This is wrong. The client immediately responds with an ACK 2 (it is not delayed), not an
ACK 6, since it wasn’t expecting byte 5 to arrive. It appears a data segment was lost. We show this
with a dashed line in Figure 19.8.

How do we know this lost segment contained bytes 2, 3, and 4, along with an ACK 3? The next byte
we’re expecting is byte number 2, as announced by segment 5. (Whenever TCP receives out-of-order
data beyond the next expected sequence number, it normally responds with an acknowledgment
specifying the sequence number of the next byte it expects to receive.) Also, since the missing segment
contained bytes 2, 3, and 4, it means the server must have received segment 2, so the missing segment
must have specified an ACK 3 (the sequence number of the next byte the server is expecting to
receive.) Finally, notice that the retransmission, segment 6, contains data from the missing segment
and segment 4. This is called repacketization, and we’ll discuss it more in Section 21.11.

265

Returning to our discussion of disabling the Nagle algorithm, we can see the 3 bytes of the next special
function key that we type is sent as three individual segments (8, 9, and 10). This time the server
echoes the byte in segment 8 first (segment II), and then echoes the bytes in segments 9 and 10
(segment 12).

What we’ve seen in this example is that the default use of the Nagle algorithm can cause additional
delays when multibyte keystrokes are entered while running an interactive application across a WAN.
We return to the topic of timeout and retransmission in Chapter 21.

Figure 19.8 Time line for Figure 19.7 (Nagle algorithm disabled).

266

19.5 Window Size Advertisements
In Figure 19.4 we see that slip advertises a window of 4096 bytes and vangogh advertises a
window of 8192 bytes. Most segments in this figure contain one of these two values.

Segment 5, however, advertises a window of 4095 bytes. This means there is still 1 byte in the TCP
buffer for the application (the Rlogin client) to read. Similarly, the next segment from the client
advertises a window of 4094 bytes, meaning there are 2 bytes still to be read.

The server normally advertises a window of 8192 bytes, because the server’s TCP has nothing to send
until the Rlogin server reads the received data and echoes it. The data from the server is sent after the
Rlogin server has read its input from the client.

The client TCP, on the other hand, often has data to send when the ACK arrives, since it’s buffering
the received characters just waiting for the ACK. When the client TCP sends the buffered data, the
Rlogin client has not had a chance to read the data received from the server, so the client’s advertised
window is less than 4096.

19.6 Summary
Interactive data is normally transmitted in segments smaller than the maximum segment size. With
Rlogin a single byte of data is normally sent from the client to the server. Telnet allows for the input to
be sent one line at a time, but most implementations today still send single characters of input.

Delayed acknowledgments are used by the receiver of these small segments to see if the
acknowledgment can be piggybacked along with data going back to the sender. This often reduces the
number of segments, especially for an Rlogin session, where the server is echoing the characters typed
at the client.

On slower WANs the Nagle algorithm is often used to reduce the number of these small segments.
This algorithm limits the sender to a single small packet of unacknowledged data at any time. But
there are times when the Nagle algorithm needs to be disabled, and we showed an example of this.

Exercises
19.1 Consider a TCP client application that writes a small application header (8 bytes) followed by a
small request (12 bytes). It then waits for a reply from the server. What happens if the request is sent
using two writes (8 bytes, then 12 bytes) versus a single write of 20 bytes?

19.2 In Figure 19.4 we are running tcpdump on the router sun. This means the data in the arrows from
the right to the left still have to go through bsdi, and the data in the arrows from the left to the right
have already come through bsdi. When we see a segment going to slip, followed by a segment
coming from slip, the time differences between the two are: 34.8, 26.7, 30.1, 28.1, 29.9, and 35.3
ms. Given that there are two links between sun and slip (an Ethernet and a 9600 bits/sec CSLIP
link), do these time differences make sense? (Hint: Reread Section 2.10.)

19.3 Compare the time required to send a special function key and have it acknowledged using the
Nagle algorithm (Figure 19.6) and with the algorithm disabled (Figure 19.8).

267

TCP Bulk Data Flow
20.1 Introduction
In Chapter 15 we saw that TFTP uses a stop-and-wait protocol. The sender of a data block required an
acknowledgment for that block before the next block was sent. In this chapter we’ll see that TCP uses
a different form of flow control called a sliding window protocol. It allows the sender to transmit
multiple packets before it stops and waits for an acknowledgment. This leads to faster data transfer,
since the sender doesn’t have to stop and wait for an acknowledgment each time a packet is sent.

We also look at TCP’s PUSH flag, something we’ve seen in many of the previous examples. We also
look at slow start, the technique used by TCP for getting the flow of data established on a connection,
and then we examine bulk data throughput.

20.2 Normal Data Flow
Let’s start with a one-way transfer of 8192 bytes from the host svr4 to the host bsdi. We run our
sock program on bsdi as the server:

bsdi % sock -i -s 7777

The -i and -s flags tell the program to run as a "sink" server (read from the network and discard the
data), and the server’s port number is specified as 7777. The corresponding client is then run as:

svr4 % sock -i -n8 bsdi 7777

This causes the client to perform eight 1024-byte writes to the network. Figure 20.1 shows the time
line for this exchange. We have left the first three segments in the output to show the MSS values for
each end.

268

Figure 20.1 Transfer of 8192 bytes from svr4 to bsdi.

The sender transmits three data segments (4-6) first. The next segment (7) acknowledges the first two
data segments only. We know this because the acknowledged sequence number is 2049, not 3073.

Segment 7 specifies an ACK of 2049 and not 3073 for the following reason. When a packet arrives it
is initially processed by the device driver’s interrupt service routine and then placed onto IP’s input
queue. The three segments 4, 5, and 6 arrive one after the other and are placed onto IP’s input queue in
the received order. IP will pass them to TCP in the same order. When TCP processes segment 4, the
connection is marked to generate a delayed ACK. TCP processes the next segment (5) and since TCP
now has two outstanding segments to ACK, the ACK of 2049 is generated (segment 7), and the
delayed ACK flag for this connection is turned off. TCP processes the next input segment (6) and the
connection is again marked for a delayed ACK. Before segment 9 arrives, however, it appears the
delayed ACK timer goes off, and the ACK of 3073 (segment 8) is generated. Segment 8 advertises a
window of 3072 bytes, implying that there are still 1024 bytes of data in the TCP receive buffer that

269

the application has not read.

Segments 11-16 show the "ACK every other segment" strategy that is common. Segments 11, 12, and
13 arrive and are placed on IP’s input queue. When segment 11 is processed by TCP the connection is
marked for a delayed ACK. When segment 12 is processed, an ACK is generated (segment 14) for
segments 11 and 12, and the delayed ACK flag for this connection is turned off. Segment 13 causes
the connection to be marked again for a delayed ACK but before the timer goes off, segment 15 is
processed, causing the ACK (segment 16) to be sent immediately.

It is important to notice that the ACK in segments 7, 14, and 16 acknowledge two received segments.
With TCP’s sliding-window protocol the receiver does not have to acknowledge every received
packet. With TCP, the ACKs are cumulative-they acknowledge that the receiver has correctly received
all bytes up through the acknowledged sequence number minus one. In this example three of the
ACKs acknowledge 2048 bytes of data and two acknowledge 1024 bytes of data. (This ignores the
ACKs in the connection establishment and termination.)

What we are watching with tcpdump are the dynamics of TCP in action. The ordering of the packets
that we see on the wire depends on many factors, most of which we have no control over: the sending
TCP implementation, the receiving TCP implementation, the reading of data by the receiving process
(which depends on the process scheduling by the operating system), and the dynamics of the network
(i.e., Ethernet collisions and backoffs). There is no single correct way for two TCPs to exchange a
given amount of data.

To show how things can change. Figure 20.2 shows another time line for the same exchange of data
between the same two hosts, captured a few minutes after the one in Figure 20.1.

270

Figure 20.2 Another transfer of 8192 bytes from svr4 to bsdi.

A few things have changed. This time the receiver does not send an ACK of 3073; instead it waits and
sends the ACK of 4097. The receiver sends only four ACKs (segments 7, 10, 12, and 15): three of
these are for 2048 bytes and one for 1024 bytes. The ACK of the final 1024 bytes of data appears in
segment 17, along with the ACK of the FIN. (Compare segment 17 in this figure with segments 16 and
18 in Figure 20.1.)

Fast Sender, Slow Receiver
Figure 20.3 shows another time line, this time from a fast sender (a Spare) to a slow receiver (an
80386 with a slow Ethernet card). The dynamics are different again.

271

Figure 20.3 Sending 8192 bytes from a fast sender to a slow receiver.

The sender transmits four back-to-back data segments (4-7) to fill the receiver’s window. The sender
then stops and waits for an ACK. The receiver sends the ACK (segment 8) but the advertised window
is 0. This means the receiver has all the data, but it’s all in the receiver’s TCP buffers, because the
application hasn’t had a chance to read the data. Another ACK (called a window update) is sent 17.4
ms later, announcing that the receiver can now receive another 4096 bytes. Although this looks like an
ACK, it is called a window update because it does not acknowledge any new data, it just advances the
right edge of the window.

The sender transmits its final four segments (10-13), again filling the receiver’s window. Notice that
segment 13 contains two flag bits: PUSH and FIN. This is followed by another two ACKs from the
receiver. Both of these acknowledge the final 4096 bytes of data (bytes 4097 through 8192) and the
FIN (numbered 8193).

20.3 Sliding Windows
The sliding window protocol that we observed in the previous section can be visualized as shown in
Figure 20.4.

272

Figure 20.4 Visualization of TCP sliding window.

In this figure we have numbered the bytes 1 through 11. The window advertised by the receiver is
called the offered window and covers bytes 4 through 9, meaning that the receiver has acknowledged
all bytes up through and including number 3, and has advertised a window size of 6. Recall from
Chapter 17 that the window size is relative to the acknowledged sequence number. The sender
computes its usable window, which is how much data it can send immediately

Over time this sliding window moves to the right, as the receiver acknowledges data. The relative
motion of the two ends of the window increases or decreases the size of the window. Three terms are
used to describe the movement of the right and left edges of the window.

1. The window closes as the left edge advances to the right. This happens when data is sent and
acknowledged.
2. The window opens when the right edge moves to the right, allowing more data to be sent. This
happens when the receiving process on the other end reads acknowledged data, freeing up space
in its TCP receive buffer.
3. The window shrinks when the right edge moves to the left. The Host Requirements RFC strongly
discourages this, but TCP must be able to cope with a peer that does this. Section 22.3 shows an
example when one side would like to shrink the window by moving the right edge to the left, but
cannot.

Figure 20.5 shows these three terms. The left edge of the window cannot move to the left, because this
edge is controlled by the acknowledgment number received from the other end. If an ACK were
received that implied moving the left edge to the left, it is a duplicate ACK, and discarded.

Figure 20.5 Movement of window edges.

If the left edge reaches the right edge, it is called a zero window. This stops the sender from
transmitting any data.

An Example
Figure 20.6 shows the dynamics of TCP’s sliding window protocol for the data transfer in Figure 20.1.

273

Figure 20.6 Sliding window protocol for Figure 20.1.

There are numerous points that we can summarize using this figure as an example.

1. The sender does not have to transmit a full window’s worth of data.
2. One segment from the receiver acknowledges data and slides the window to the right. This is
because the window size is relative to the acknowledged sequence number.
3. The size of the window can decrease, as shown by the change from segment 7 to segment 8, but
the right edge of the window must not move leftward.
4. The receiver does not have to wait for the window to fill before sending an ACK. We saw earlier
that many implementations send an ACK for every two segments that are received.

We’ll see more examples of the dynamics of the sliding window protocol in later examples.

20.4 Window Size
The size of the window offered by the receiver can usually be controlled by the receiving process. This
can affect the TCP performance.

4.2BSD defaulted the send buffer and receive buffer to 2048 bytes each. With 4.3BSD both were increased to 4096 bytes.
As we can see from all the examples so far in this text, SunOS 4.1.3, BSD/386, and SVR4 still use this 4096-byte default.
Other systems, such as Solaris 2.2,4.4BSD, and AIX 3.2, use larger default buffer sizes, such as 8192 or 16384 bytes.

The sockets API allows a process to set the sizes of the send buffer and the receive buffer. The size of the receive buffer is
the maximum size of the advertised window for that connection. Some applications change the socket buffer sizes to
increase performance.

274

[Mogul 1993] shows some results for file transfer between two workstations on an Ethernet, with
varying sizes for the transmit buffer and receive buffer. (For a one-way flow of data such as file
transfer, it is the size of the transmit buffer on the sending side and the size of the receive buffer on the
receiving side that matters.) The common default of 4096 bytes for both is not optimal for an Ethernet.
An approximate 40% increase in throughput is seen by just increasing both buffers to 16384 bytes.
Similar results are shown in [Papadopoulos and Parulkar 1993].

In Section 20.7 we’ll see how to calculate the minimum buffer size, given the bandwidth of the
communication media and the round-trip time between the two ends.

An Example
We can control the sizes of these buffers with our sock program. We invoke the server as:

bsdi % sock -i -a -R6144 5555

which sets the size of the receive buffer (-R option) to 6144 bytes. We then start the client on the host
sun and have it perform one write of 8192 bytes:

sun % sock -i -nl -w8192 bsdi 5555

Figure 20.7 shows the results.

275

Figure 20.7 Data transfer with receiver offering a window size of 6144 bytes.

First notice that the receiver’s window size is offered as 6144 bytes in segment 2. Because of this
larger window, the client sends six segments immediately (segments 4-9), and then stops. Segment 10
acknowledges all the data (bytes 1 through 6144) but offers a window of only 2048, probably because
the receiving application hasn’t had a chance to read more than 2048 bytes. Segments 11 and 12
complete the data transfer from the client, and this final data segment also carries the FIN flag.

Segment 13 contains the same acknowledgment sequence number as segment 10, but advertises a
larger window. Segment 14 acknowledges the final 2048 bytes of data and the FIN, and segments 15
and 16 just advertise a larger window. Segments 17 and 18 complete the normal close.

20.5 PUSH Flag
We’ve seen the PUSH flag in every one our TCP examples, but we’ve never described its use. It’s a
notification from the sender to the receiver for the receiver to pass all the data that it has to the
receiving process. This data would consist of whatever is in the segment with the PUSH flag, along
with any other data the receiving TCP has collected for the receiving process.

276

In the original TCP specification, it was assumed that the programming interface would allow the
sending process to tell its TCP when to set the PUSH flag. In an interactive application, for example,
when the client sent a command to the server, the client would set the PUSH flag and wait for the
server’s response. (In Exercise 19.1 we could imagine the client setting the PUSH flag when the
12-byte request is written.) By allowing the client application to tell its TCP to set the flag, it was a
notification to the client’s TCP that the client process didn’t want the data to hang around in the TCP
buffer, waiting for additional data, before sending a segment to the server. Similarly, when the server’s
TCP received the segment with the PUSH flag, it was a notification to pass the data to the server
process and not wait to see if any additional data arrives.

Today, however, most APIs don’t provide a way for the application to tell its TCP to set the PUSH
flag. Indeed, many implementors feel the need for the PUSH flag is outdated, and a good TCP
implementation can determine when to set the flag by itself.

Most Berkeley-derived implementations automatically set the PUSH flag if the data in the segment
being sent empties the send buffer. This means we normally see the PUSH flag set for each application
write, because data is usually sent when it’s written.

A comment in the code indicates this algorithm is to please those implementations that only pass received data to the
application when a buffer fills or a segment is received with the PUSH flag.

It is not possible using the sockets API to tell TCP to turn on the PUSH flag or to tell whether the PUSH flag was set in
received data.

Berkeley-derived implementations ignore a received PUSH flag because they normally never delay the
delivery of received data to the application.

Examples
In Figure 20.1 we see the PUSH flag turned on for all eight data segments (4-6, 9, 11-13, and 15). This
is because the client did eight writes of 1024 bytes, and each write emptied the send buffer.

Look again at Figure 20.7. We expect the PUSH flag to be set on segment 12, since that is the final
data segment. Why was the PUSH flag set on segment 7, when the sender knew there were still more
bytes to send? The reason is that the size of the sender’s send buffer is 4096 bytes, even though we
specified a single write of 8192 bytes.

Another point to note in Figure 20.7 concerns the three consecutive ACKs, segments 14, 15, and 16.
We saw two consecutive ACKs in Figure 20.3, but that was because the receiver had advertised a
window of 0 (stopping the sender) so when the window opened up, another ACK was required, with
the nonzero window, to restart the sender. In Figure 20.7, however, the window never reaches 0.
Nevertheless, when the size of the window increases by 2048 bytes, another ACK is sent (segments 15
and 16) to provide this window update to the other end. (These two window updates in segments 15
and 16 are not needed, since the FIN has been received from the other end, preventing it from sending
any more data.) Many implementations send this window update if the window increases by either two
maximum sized segments (2048 bytes in this example, with an MSS of 1024) or 50% of the maximum
possible window (2048 bytes in this example, with a maximum window of 4096). We’ll see this again
in Section 22.3 when we examine the silly window syndrome in detail.

As another example of the PUSH flag, look again at Figure 20.3. The reason we see the flag on for the
first four data segments (4-7) is because each one caused a segment to be generated by TCP and
passed to the IP layer. But then TCP had to stop, waiting for an ACK to move the 4096-byte window.
While waiting for the ACK, TCP takes the final 4096 bytes of data from the application. When the

277

window opens up (segment 9) the sending TCP knows it has four segments that it can send
immediately, so it only turns on the PUSH flag for the final segment (13).

20.6 Slow Start
In all the examples we’ve seen so far in this chapter, the sender starts off by injecting multiple
segments into the network, up to the window size advertised by the receiver. While this is OK when
the two hosts are on the same LAN, if there are routers and slower links between the sender and the
receiver, problems can arise. Some intermediate router must queue the packets, and it’s possible for
that router to run out of space. [Jacobson 1988] shows how this naive approach can reduce the
throughput of a TCP connection drastically.

TCP is now required to support an algorithm called slow start. It operates by observing that the rate at
which new packets should be injected into the network is the rate at which the acknowledgments are
returned by the other end.

Slow start adds another window to the sender’s TCP: the congestion window, called cwnd. When a
new connection is established with a host on another network, the congestion window is initialized to
one segment (i.e., the segment size announced by the other end). Each time an ACK is received, the
congestion window is increased by one segment, (cwnd is maintained in bytes, but slow start always
increments it by the segment size.) The sender can transmit up to the minimum of the congestion
window and the advertised window. The congestion window is flow control imposed by the sender,
while the advertised window is flow control imposed by the receiver.

The sender starts by transmitting one segment and waiting for its ACK. When that ACK is received,
the congestion window is incremented from one to two, and two segments can be sent. When each of
those two segments is acknowledged, the congestion window is increased to four. This provides an
exponential increase.

At some point the capacity of the internet can be reached, and an intermediate router will start
discarding packets. This tells the sender that its congestion window has gotten too large. When we talk
about TCP’s timeout and retransmission algorithms in the next chapter, we’ll see how this is handled,
and what happens to the congestion window. For now, let’s watch slow start in action.

An Example
Figure 20.8 shows data being sent from the host sun to the host vangogh.cs.berkeley.edu.
The data traverses a slow SLIP link, which should be the bottleneck. (We have removed the
connection establishment from this time line.)

278

Figure 20.8 Example of slow start.

We see the sender transmit one segment with 512 bytes of data and then wait for its ACK. The ACK is
received 716 ms later, which is an indicator of the round-trip time. The congestion window is then
increased to two segments, and two segments are sent. When the ACK in segment 5 is received, the
congestion window is increased to three segments. Although three more could be sent, only two are
sent before another ACK is received.

We’ll return to slow start in Section 21.6 and see how it’s normally implemented with another
technique called congestion avoidance.

20.7 Bulk Data Throughput
Let’s look at the interaction of the window size, the windowed flow control, and slow start on the
throughput of a TCP connection carrying bulk data.

Figure 20.9 shows the steps over time of a connection between a sender on the left and a receiver on
the right. Sixteen units of time are shown. We show only discrete units of time in this figure, for
simplicity. We show segments carrying data going from the left to right in the top half of each picture,
numbered 1, 2, 3, and so on. The ACKs go in the other direction in the bottom half of each picture. We

279

draw the ACKs smaller, and show the segment number being acknowledged.

Figure 20.9 Times 0-15 for bulk data throughput example.

At time 0 the sender transmits one segment. Since the sender is in slow start (its congestion window is
one segment), it must wait for the acknowledgment of this segment before continuing.

At times 1, 2, and 3 the segment moves one unit of time to the right. At time 4 the receiver reads the
segment and generates the acknowledgment. At times 5, 6, and 7 the ACK moves to the left one unit,
back to the sender. We have a round-trip time (RTT) of 8 units of time.

We have purposely drawn the ACK segment smaller than the data segment, since it’s normally just an
IP header and a TCP header. We’re showing only a unidirectional flow of data here. Also, we assume
that the ACK moves at the same speed as the data segment, which isn’t always true.

In general the time to send a packet depends on two factors: a propagation delay (caused by the finite speed of light,
latencies in transmission equipment, etc.) and a transmission delay that depends on the speed of the media (how many bits
per second the media can transmit). For a given path between two nodes the propagation delay is fixed while the
transmission delay depends on the packet size. At lower speeds the transmission delay dominates (e.g., Exercise 7.2 where
we didn’t even consider the propagation delay), whereas at gigabit speeds the propagation delay dominates (e.g.. Figure
24.6).

280

When the sender receives the ACK it can transmit two more segments (which we’ve numbered 2 and
3), at times 8 and 9. Its congestion window is now two segments. These two segments move right
toward the receiver, where the ACKs are generated at times 12 and 13. The spacing of the ACKs
returned to the sender is identical to the spacing of the data segments. This is called the self-clocking
behavior of TCP. Since the receiver can only generate ACKs when the data arrives, the spacing of the
ACKs at the sender identifies the arrival rate of the data at the receiver. (In actuality, however,
queueing on the return path can change the arrival rate of the ACKs.)

Figure 20.10 shows the next 16 time units. The arrival of the two ACKs increases the congestion
window from two to four segments, and these four segments are sent at times 16-19. The first of the
ACKs returns at time 23. The four ACKs increase the congestion window from four to eight segments,
and these eight segments are transmitted at times 24-31.

Figure 20.10 Times 16-31 for bulk data throughput example.

At time 31, and at all successive times, the pipe between the sender and receiver is full. It cannot hold
any more data, regardless of the congestion window or the window advertised by the receiver. Each
unit of time a segment is removed from the network by the receiver, and another is placed into the
network by the sender. However many data segments fill the pipe, there are an equal number of ACKs

281

making the return trip. This is the ideal steady state of the connection.

Bandwidth-Delay Product
We can now answer the question: how big should the window be? In our example, the sender needs to
have eight segments outstanding and unacknowledged at any time, for maximum throughput. The
receiver’s advertised window must be that large, since that limits how much the sender can transmit.
We can calculate the capacity of the pipe as

capacity (bits) = bandwidth (bits/sec) x round-trip time (sec)

This is normally called the bandwidth-delay product. This value can vary widely, depending on the
network speed and the RTT between the two ends. For example, a Tl telephone line (1,544,000
bits/sec) across the United States (about a 60-ms RTT) gives a bandwidth-delay product of 11,580
bytes. This is reasonable in terms of the buffer sizes we talked about in Section 20.4, but a T3
telephone line (45,000,000 bits/sec) across the United States gives a bandwidth-delay product of
337,500 bytes, which is bigger than the maximum allowable TCP window advertisement (65535
bytes). We describe the new TCP window scale option in Section 24.4 that gets around this current
limitation of TCP.

The value 1,544,000 bits/sec for a Tl phone line is the raw bit rate. The data rate is actually 1,536,000 bits/sec, since 1 bit
in 193 is used for framing. The raw bit rate of a T3 phone line is actually 44,736,000 bits/sec, and the data rate can reach
44,210,000 bits/sec. For our discussion we’ll use 1.544 Mbits/sec and 45 Mbits/sec.

Either the bandwidth or the delay can affect the capacity of the pipe between the sender and receiver.
In Figure 20.11 we show graphically how a doubling of the RTT-doubles the capacity of the pipe.

Figure 20.11 Doubling the RTT doubles the capacity of the pipe.

In the lower illustration of Figure 20.11, with the longer RTT, the pipe can hold eight segments,
instead of four.

Similarly, Figure 20.12 shows that doubling the bandwidth also doubles the capacity of the pipe.

Figure 20.12 Doubling the bandwidth doubles the capacity of the pipe.

282

In the lower illustration of Figure 20.12, we assume that the network speed has doubled, allowing us to
send four segments in half the time as in the top picture. Again, the capacity of the pipe has doubled.
(We assume that the segments in the top half of this figure have the same area, that is the same number
of bits, as the segments in the bottom half.)

Congestion
Congestion can occur when data arrives on a big pipe (a fast LAN) and gets sent out a smaller pipe (a
slower WAN). Congestion can also occur when multiple input streams arrive at a router whose output
capacity is less than the sum of the inputs.

Figure 20.13 shows a typical scenario with a big pipe feeding a smaller pipe. We say this is typical
because most hosts are connected to LANs, with an attached router that is connected to a slower
WAN. (Again, we are assuming the areas of all the data segments (9-20) in the top half of the figure
are all the same, and the areas of all the acknowledgments in the bottom half are all the same.)

Figure 20.13 Congestion caused by a bigger pipe feeding a smaller pipe.

In this figure we have labeled the router R1 as the "bottleneck," because it is the congestion point. It
can receive packets from the LAN on its left faster than they can be sent out the WAN on its right.
(Commonly R1 and R3 are the same router, as are R2 and R4, but that’s not required; asymmetrical
paths can occur.) When router R2 puts the received packets onto the LAN on its right, they maintain
the same spacing as they did on the WAN on its left, even though the bandwidth of the LAN is higher.
Similarly, the spacing of the ACKs on their way back is the same as the spacing of the slowest link in
the path.

In Figure 20.13 we have assumed that the sender did not use slow start, and sent the segments we’ve
numbered 1-20 as fast as the LAN could take them. (This assumes the receiving host advertised a
window of at least 20 segments.) The spacing of the ACKs will correspond to the bandwidth of the
slowest link, as we show. We are assuming the bottleneck router has adequate buffering for all 20
segments. This is not guaranteed, and can lead to that router discarding packets. We’ll see how to
avoid this when we talk about congestion avoidance in Section 21.6.

20.8 Urgent Mode
TCP provides what it calls urgent mode, allowing one end to tell the other end that "urgent data" of
some form has been placed into the normal stream of data. The other end is notified that this urgent
data has been placed into the data stream, and it’s up to the receiving end to decide what to do.

283

The notification from one end to the other that urgent data exists in the data stream is done by setting
two fields in the TCP header (Figure 17.2). The URG bit is turned on and the 16-bit urgent pointer is
set to a positive offset that must be added to the sequence number field in the TCP header to obtain the
sequence number of the last byte of urgent data.

There is continuing debate about whether the urgent pointer points to the last byte of urgent data, or to the byte following
the last byte of urgent data. The original TCP specification gave both interpretations but the Host Requirements RFC
identifies which is correct: the urgent pointer points to the last byte of urgent data.

The problem, however, is that most implementations (i.e., the Berkeley-derived implementations) continue to use the
wrong interpretation. An implementation that follows the specification in the Host Requirements RFC might be compliant,
but might not communicate correctly with most other hosts.

TCP must inform the receiving process when an urgent pointer is received and one was not already
pending on the connection, or if the urgent pointer advances in the data stream. The receiving
application can then read the data stream and must be able to tell when the urgent pointer is
encountered. As long as data exists from the receiver’s current read position until the urgent pointer,
the application is considered to be in an "urgent mode." After the urgent pointer is passed, the
application returns to its normal mode.

TCP itself says little more about urgent data. There is no way to specify where the urgent data starts in
the data stream. The only information sent across the connection by TCP is that urgent mode has
begun (the URG bit in the TCP header) and the pointer to the last byte of urgent data. Everything else
is left to the application.

Unfortunately many implementations incorrectly call TCP’s urgent mode out-of-band data. If an
application really wants a separate out-of-band channel, a second TCP connection is the easiest way to
accomplish this. (Some transport layers do provide what most people consider true out-of-band data: a
logically separate data path using the same connection as the normal data path. This is not what TCP
provides.)

The confusion between TCP’s urgent mode and out-of-band data is also because the predominant programming interface,
the sockets API, maps TCP’s urgent mode into what sockets calls out-of-band data.

What is urgent mode used for? The two most commonly used applications are Tel-net and Rlogin,
when the interactive user types the interrupt key, and we show examples of this use of urgent mode in
Chapter 26. Another is FTP, when the interactive user aborts a file transfer, and we show an example
of this in Chapter 27.

Telnet and Rlogin use urgent mode from the server to the client because it’s possible for this direction
of data flow to be stopped by the client TCP (i.e., it advertises a window of 0). But if the server
process enters urgent mode, the server TCP immediately sends the urgent pointer and the URG flag,
even though it can’t send any data. When the client TCP receives this notification, it in turn notifies
the client process, so the client can read its input from the server, to open the window, and let the data
flow.

What happens if the sender enters urgent mode multiple times before the receiver processes all the data
up through the first urgent pointer? The urgent pointer just advances in the data stream, and its
previous position at the receiver is lost. There is only one urgent pointer at the receiver and its value is
overwritten when a new value for the urgent pointer arrives from the other end. This means if the
contents of the data stream that are written by the sender when it enters urgent mode are important to
the receiver, these data bytes must be specially marked (somehow) by the sender. We’ll see that Telnet
marks all of its command bytes in the data stream by prefixing them with a byte of 255.

284

An Example
Let’s watch how TCP sends urgent data, even when the receiver’s window is closed. We’ll start our
sock program on the host bsdi and have it pause for 10 seconds after the connection is established
(the -P option), before it reads from the network. This lets the other end fill the send window.

bsdi % sock -i -s -P10 5555

We then start the client on the host sun telling it to use a send buffer of 8192 bytes (-S option) and
perform six 1024-byte writes to the network (-n option). We also specify -U5 telling it to write 1 byte
of data and enter urgent mode before writing the fifth buffer to the network. We specify the verbose
flag to see the order of the writes:

sun % sock -v -i -n6 -S8192 -U5 bsdi 5555
connected on 140.252.13.33.1305 to 140.252.13.35.5555
SO_SNDBUF = 8192
TCP_MAXSEG = 1024
wrote 1024 bytes
wrote 1024 bytes
wrote 1024 bytes
wrote 1024 bytes
wrote 1 byte of urgent data
wrote 1024 bytes
wrote 1024 bytes

We set the send buffer size to 8192 bytes, to let the sending application immediately write all of its
data. Figure 20.14 shows the tcpdump output for this exchange. (We have removed the connection
establishment.) Lines 1-5 show the sender filling the receiver’s window with four 1024-byte segments.
The sender is then stopped because the receiver’s window is full. (The ACK on line 4 acknowledges
data, but does not move the right edge of the window.)

sun.1305 > bsdi.5555: P 1:1025(1024) ack 1 win
1 0.0
4096
0.073743 sun.1305 > bsdi.5555: P 1025:2049(1024) ack 1
2
(0.0737) win 4096
0.096969 sun.1305 > bsdi.5555: P 2049:3073(1024) ack 1
3
(0.0232) win 4096
0.157514
4 bsdi.5555 > sun.1305: . ack 3073 win 1024
(0.0605)
0.164267 sun.1305 > bsdi.5555: P 3073:4097(1024) ack 1
5
(0.0068) win 4096
0.167961 sun.1305 > bsdi.5555: . ack 1 win 4096 urg
6
(0.0037) 4098
7
(0.0040) 4098

285

8
(0.0042) 4098
9
(0.0042) 4098
10
(0.0004) 4098
0.367533
11 bsdi.5555 > sun.1305: . ack 4097 win 0
(0.1868)
12
(0.0009) 4098
9.829712
13 bsdi.5555 > sun.1305: . ack 4097 win 2048
(9.4612)
9.831578 sun.1305 > bsdi.5555: . 4097:5121(1024) ack 1
14
(0.0019) win 4096 urg 4098
9.833303 sun.1305 > bsdi.5555: . 5121:6145(1024) ack 1
15
(0.0017) win 4096
9.835089
16 bsdi.5555 > sun.1305: . ack 4097 win 4096
(0.0018)
9.835913 sun.1305 > bsdi.5555: FP 6145:6146(1) ack 1
17
(0.0008) win 4096
9.840264
18 bsdi.5555 > sun.1305: . ack 6147 win 2048
(0.0044)
9.842386
19 bsdi.5555 > sun.1305: . ack 6147 win 4096
(0.0021)
9.843622 bsdi.5555 > sun.1305: F 1:1(0) ack 6147 win
20
(0.0012) 4096
9.844320
21 sun.1305 > bsdi.5555: . ack 2 win 4096
(0.0007)

Figure 20.14 tcpdump output for TCP urgent mode.

After the fourth application write of normal data, the application writes 1 byte of data and enters
urgent mode. Line 6 is the result of this application write. The urgent pointer is set to 4098. The urgent
pointer is sent with the URG flag even though the sender cannot send any data.

Five of these ACKs are sent in about 13 ms (lines 6-10). The first is sent when the application writes 1
byte and enters urgent mode. The next two are sent when the application does the final two writes of
1024 bytes. (Even though TCP can’t send these 2048 bytes of data, each time the application performs
a write, the TCP output function is called, and when it sees that urgent mode has been entered, sends
another urgent notification.) The fourth of these ACKs occurs when the application closes its end of
the connection. (The TCP output function is again called.) The sending application terminates
milliseconds after it starts-before the receiving application has issued its first write. TCP queues all the

286

data and sends it when it can. (This is why we specified a send buffer size of 8192-so all the data can
fit in the buffer.) The fifth of these ACKs is probably generated by the reception of the ACK on line 4.
The sending TCP has probably already queued its fourth segment for output (line 5) before this ACK
arrives. The receipt of this ACK from the other end also causes the TCP output routine to be called.

The receiver then acknowledges the final 1024 bytes of data (line 11) but also advertises a window of
0. The sender responds with another segment containing the urgent notification.

The receiver advertises a window of 2048 bytes in line 13, when the application wakes up and reads
some of the data from the receive buffer. The next two 1024-byte segments are sent (lines 14 and 15).
The first segment has the urgent notification set, since the urgent pointer is within this segment. The
second segment has turned the urgent notification off.

When the receiver opens the window again (line 16) the sender transmits the final byte of data
(numbered 6145) and also initiates the normal connection termination.

Figure 20.15 shows the sequence numbers of the 6145 bytes of data that are sent. We see that the
sequence number of the byte written when urgent mode was entered is 4097, but the value of the
urgent pointer in Figure 20.14 is 4098. This confirms that this implementation (SunOS 4.1.3) sets the
urgent pointer to 1 byte beyond the last byte of urgent data.

Figure 20.15 Application writes and TCP segments for urgent mode example.

This figure also lets us see how TCP repacketizes the data that the application wrote. The single byte
that was output when urgent mode was entered is sent along with the next 1023 bytes of data in the
buffer. The next segment also contains 1024 bytes of data, and the final segment contains 1 byte of
data.

20.9 Summary
As we said early in the chapter, there is no single way to exchange bulk data using TCP. It is a
dynamic process that depends on many factors, some of which we can control (e.g., send and receive
buffer sizes) and some of which we have no control over (e.g., network congestion, implementation
features). In this chapter we’ve examined many TCP transfers, explaining all the characteristics and
algorithms that we could see.

Fundamental to the efficient transfer of bulk data is TCP’s sliding window protocol. We then looked at
what it takes for TCP to get the fastest transfer possible by keeping the pipe between the sender and
receiver full. We measured the capacity of this pipe as the bandwidth-delay product, and saw the
relationship between this and the window size. We return to this concept in Section 24.8 when we look
at TCP performance.

We also looked at TCP’s PUSH flag, since we’ll always see it in trace output, but we have no control
over its setting. The final topic was TCP’s urgent data, which is often mistakenly called "out-of-band
data." TCP’s urgent mode is just a notification from the sender to the receiver that urgent data has been
sent, along with the sequence number of the final byte of urgent data. The programming interface for

287

the application to use with urgent data is often less than optimal, which leads to much confusion.

Exercises
20.1 In Figure 20.6 we could have shown a byte numbered 0 and a byte numbered 8193. What do
these 2 bytes designate?

20.2 Look ahead to Figure 22.1 and explain the setting of the PUSH flag by the host bsdi.

20.3 In a Usenet posting someone complained about a throughput of 120,000 bits/sec on a 256,000
bits/sec link with a 128-ms delay between the United States and Japan (47% utilization), and a
throughput of 33,000 bits/sec when the link was routed over a satellite (13% utilization). What does
the window size appear to be for both cases? (Assume a 500-ms delay for the satellite link.) How big
should the window be for the satellite link?

20.4 If the API provided a way for a sending application to tell its TCP to turn on the PUSH flag, and
a way for the receiver to tell if the PUSH flag was on in a received segment, could the flag then be
used as a record marker?

20.5 In Figure 20.3 why aren’t segments 15 and 16 combined?

20.6 In Figure 20.13 we assume that the ACKs come back nicely spaced, corresponding to the spacing
of the data segments. What happens if the ACKs are queued somewhere on the return path, causing a
bunch of them to arrive at the same time at the sender?

288

TCP Timeout and Retransmission
21.1 Introduction
TCP provides a reliable transport layer. One of the ways it provides reliability is for each end to
acknowledge the data it receives from. the other end. But data segments and acknowledgments can get
lost. TCP handles this by setting a timeout when it sends data, and if the data isn’t acknowledged when
the timeout expires, it retransmits the data. A critical element of any implementation is the timeout and
retransmission strategy. How is the timeout interval determined, and how frequently does a
retransmission occur?

We’ve already seen two examples of timeout and retransmission: (1) In the ICMP port unreachable
example in Section 6.5 we saw the TFTP client using UDP employing a simple (and poor) timeout and
retransmission strategy: it assumed 5 seconds was an adequate timeout period and retransmitted every
5 seconds. (2) In the ARP example to a nonexistent host (Section 4.5), we saw that when TCP tried to
establish the connection it retransmitted its SYN using a longer delay between each retransmission.
TCP manages four different timers for each connection.

1. A retransmission timer is used when expecting an acknowledgment from the other end. This
chapter looks at this timer in detail, along with related issues such as congestion avoidance.
2. A persist timer keeps window size information flowing even if the other end closes its receive
window. Chapter 22 describes this timer.
3. A keepalive timer detects when the other end on an otherwise idle connection crashes or reboots.
Chapter 23 describes this timer.
4. A 2MSL timer measures the time a connection has been in the TIME_WAIT state. We described
this state in Section 18.6.

In this chapter we start with a simple example of TCP’s timeout and retransmission and then move to a
larger example that lets us look at all the details involved in TCP’s timer management. We look at how
typical implementations measure the round-trip time of TCP segments and how TCP uses these
measurements to estimate the retransmission timeout of the next segment it transmits. We then look at
TCP’s congestion avoidance-what TCP does when packets are lost-and follow through an actual
example where packets are lost. We also look at the newer fast retransmit and fast recovery
algorithms, and see how they let TCP detect lost packets faster than waiting for a timer to expire.

21.2 Simple Timeout and Retransmission Example
Let’s first look at the retransmission strategy used by TCP. We’ll establish a connection, send some
data to verify that everything is OK, disconnect the cable, send some more data, and watch what TCP
does:

289

bsdi % telnet svr4 discard
Trying 140.252.13.34...
Connected to svr4.
Escape character is ’^]’.
Hello, world send this line normally
and hi disconnect cable before sending this line
Connection closed by foreign host. output whenTCP gives up after 9 minutes

Figure 21.1 shows the tcpdump output. (We have removed all the type-of-service information that is
set by bsdi.)

bsdi.1029 > svr4.discard: S
1 0.0 1747921409:1747921409(0)
win 4096 <mss 1024>
svr4.discard > bsdi.1029: S
0.004811 (
2 3416685569:3416685569(0)
0.0048)
ack 1747921410 win 4096 <mss 1024>
0.006441 (
3 bsdi.1029 > svr4.discard: . ack 1 win 4096
0.0016)
6.102290 ( bsdi.1029 > svr4.discard: P 1:15(14) ack 1
4
6.0958) win 4096
6.259410 (
5 svr4.discard > bsdi.1029: . ack 15 win 4096
0.1571)
24.480158 bsdi.1029 > svr4.discard: P 15:23(8) ack 1
6
(18.2207) win 4096
7
1.0136) win 4096
8
3.0001) win 4096
9
6.0002) win 4096
10
(11.9905) win 4096
11
(24.0007) win 4096
12
(48.0013) win 4096

290

13
(64.0018) win 4096
14
(64.0018) win 4096
15
(64.0018) win 4096
16
(64.0018) win 4096
17
(64.0018) win 4096
18
(63.9917) win 4096
566.488478 bsdi.1029 > svr4.discard: R 23:23(0) ack 1
19
(64.0015) win 4096

Figure 21.1 Simple example of TCP’s timeout and retransmission.

Lines 1, 2, and 3 correspond to the normal TCP connection establishment. Line 4 is the transmission
of "hello, world" (12 characters plus the carriage return and linefeed), and line 5 is its
acknowledgment. We then disconnect the Ethernet cable from svr4.

Line 6 shows "and hi" being sent. Lines 7-18 are 12 retransmissions of that segment, and line 19 is
when the sending TCP finally gives up and sends a reset.

Examine the time difference between successive retransmissions: with rounding they occur 1, 3, 6, 12,
24, 48, and then 64 seconds apart. We’ll see later in this chapter that the first timeout is actually set for
1.5 seconds after the first transmission. (The reason it occurs 1.0136 seconds after the first
transmission, and not exactly 1.5 seconds, was explained in Figure 18.7.) After this the timeout value
is doubled for each retransmission, with an upper limit of 64 seconds.

This doubling is called an exponential backoff. Compare this to the TFTP example in Section 6.5,
where every retransmission occurred 5 seconds after the previous.

The time difference between the first transmission of the packet (line 6 at time 24.480) and the reset
(line 19 at time 566.488) is about 9 minutes. Modem TCP’s are persistent when trying to send data!

On most implementations this total timeout value is not tunable. Solaris 2.2 allows the administrator to change this (the
tcp_ip_abort_interval variable in Section E.4) and its default is only 2 minutes, not the more common 9 minutes.

21.3 Round-Trip Time Measurement
Fundamental to TCP’s timeout and retransmission is the measurement of the round-trip time (RTT)
experienced on a given connection. We expect this can change over time, as routes might change and
as network traffic changes, and TCP should track these changes and modify its timeout accordingly.

291

First TCP must measure the RTT between sending a byte with a particular sequence number and
receiving an acknowledgment that covers that sequence number. Recall from the previous chapter that
normally there is not a one-to-one correspondence between data segments and ACKs. In Figure 20.1
this means that one RTT that can be measured by the sender is the time between the transmission of
segment 4 (data bytes 1-1024) and the reception of segment 7 (the ACK of bytes 1-2048), even though
this ACK is for an additional 1024 bytes. We’ll use M to denote the measured RTT.

The original TCP specification had TCP update a smoothed RTT estimator (called R) using the
low-pass filter

R <- aR + (1-a)M
where a is a smoothing factor with a recommended value of 0.9. This smoothed RTT is updated every
time a new measurement is made. Ninety percent of each new estimate is from the previous estimate
and 10% is from the new measurement.

Given this smoothed estimator, which changes as the RTT changes, RFC 793 recommended the
retransmission timeout value (RTO) be set to

RTO = Rb
where b is a delay variance factor with a recommended value of 2.

[Jacobson 1988] details the problems with this approach, basically that it can’t keep up with wide fluctuations in the RTT,
causing unnecessary retransmissions. As Jacobson notes, unnecessary retransmissions add to the network load, when the
network is already loaded. It is the network equivalent of pouring gasoline on a fire.

What’s needed is to keep track of the variance in the RTT measurements, in addition to the smoothed
RTT estimator. Calculating the RTO based on both the mean and variance provides much better
response to wide fluctuations in the round-trip times, than just calculating the RTO as a constant
multiple of the mean. Figures 5 and 6 in [Jacobson 1988] show a comparison of the RFC 793 RTO
values for some actual round-trip times, versus the RTO calculations we show below, which take into
account the variance of the round-trip times.

As described by Jacobson, the mean deviation is a good approximation to the standard deviation, but
easier to compute. (Calculating the standard deviation requires a square root.) This leads to the
following equations that are applied to each RTT measurement M.

Err = M - A
A <- A + gErr
D <- D+ h(|Err| - D)
RTO = A + 4D
where A is the smoothed RTT (an estimator of the average) and D is the smoothed mean deviation. Err
is the difference between the measured value just obtained and the current RTT estimator. Both A and
D are used to calculate the next retransmission timeout (RTO). The gain g is for the average and is set
to 1/8 (0.125). The gain for the deviation is h and is set to 0.25. The larger gain for the deviation
makes the RTO go up faster when the RTT changes.

292

[Jacobson 1988] specified 2D in the calculation of RTO, but after further research, [Jacobson 1990c] changed the value to
4D, which is what appeared in the BSD Net/1 implementation.

Jacobson specifies a way to do all these calculations using integer arithmetic, and this is the
implementation typically used. (That’s one reason g, h, and the multiplier 4 are all powers of 2, so the
operations can be done using shifts instead of multiplies and divides.)

Comparing the original method with Jacobson’s, we see that the calculations of the smoothed average
are similar (a is one minus the gain g) but a different gain is used. Also, Jacobson’s calculation of the
RTO depends on both the smoothed RTT and the smoothed mean deviation, whereas the original
method used a multiple of the smoothed RTT.

We’ll see how these estimators are initialized in the next section, when we go through an example.

Karn’s Algorithm
A problem occurs when a packet is retransmitted. Say a packet is transmitted, a timeout occurs, the
RTO is backed off as shown in Section 21.2, the packet is retransmitted with the longer RTO, and an
acknowledgment is received. Is the ACK for the first transmission or the second? This is called the
retransmission ambiguity problem.

[Karn and Partridge 1987] specify that when a timeout and retransmission occur, we cannot update the
RTT estimators when the acknowledgment for the retransmitted data finally arrives. This is because
we don’t know to which transmission the ACK corresponds. (Perhaps the first transmission was
delayed and not thrown away, or perhaps the ACK of the first transmission was delayed.)

Also, since the data was retransmitted, and the exponential backoff has been applied to the RTO, we
reuse this backed off RTO for the next transmission. Don’t calculate a new RTO until an
acknowledgment is received for a segment that was not retransmitted.

21.4 An RTT Example
We’ll use the following example throughout this chapter to examine various implementation details of
TCP’s timeout and retransmission, slow start, and congestion avoidance.

Using our sock program, 32768 bytes of data are sent from our host slip to the discard service on
the host vangogh.cs.berkeley.edu using the command:

slip % sock -D -i -n32 vangogh.cs.berkeley.edu discard

From the figure on the inside front cover, slip is connected from the 140.252.1 Ethernet by two
SLIP links, and from there across the Internet to the destination. With two 9600 bits/sec SLIP links,
we expect some measurable delays.

This command performs 32 1024-byte writes, and since the MTU between slip and bsdi is 296,
this becomes 128 segments, each with 256 bytes of user data. The total time for the transfer is about 45
seconds and we see one timeout and three retransmissions.

While this transfer was running we ran tcpdump on the host slip and captured all the segments sent
and received. Additionally we specified the -D option to turn on socket debugging (Section A.6). We
were then able to run a modified version of the trpt(8) program to print numerous variables in the
connection control block relating to the round-trip timing, slow start, and congestion avoidance.

293

Given the volume of trace output, we can’t show it all. Instead we’ll look at pieces as we proceed
through the chapter. Figure 21.2 shows the transfer of data and acknowledgments for the first 5
seconds. We have modified this output slightly from our previous display of tcpdump output.
Although we only measure the times that the packet is sent or received on the host running tcpdump,
in this figure we want to show that the packets are crossing in the network (which they are, since this
LAN connection is not like a shared Ethernet), and show when the receiving host is probably
generating the ACKs. (We have also removed all the window advertisements from this figure, slip
always advertised a window of 4096, and vangogh always advertised a window of 8192.)

[IMAGE]
Figure 21.2 Packet exchange and RTT measurement.

Also note in this figure that we have numbered the segments 1-13 and 15, in the order in which they
were sent or received on the host slip. This correlates with the tcpdump output that was collected
on this host.

Round-Trip Time Measurements
Three curly braces have been placed on the left side of the time line indicating which segments were
timed for RTT calculations. Not all data segments are timed.

Most Berkeley-derived implementations of TCP measure only one RTT value per connection at any
time. If the timer for a given connection is already in use when a data segment is transmitted, that
segment is not timed.

The timing is done by incrementing a counter every time the 500-ms TCP timer routine is invoked.
This means that a segment whose acknowledgment arrives 550 rns after the segment was sent could
end up with either a 1 tick RTT (implying 500 ms) or a 2 tick RTT (implying 1000 ms).

In addition to this tick counter for each connection, the starting sequence number of the data in the
segment is also remembered. When an acknowledgment that includes this sequence number is
received, the timer is turned off. If the data was not retransmitted when the ACK arrives, the smoothed
RTT and smoothed mean deviation are updated based on this new measurement.

The timer for the connection in Figure 21.2 is started when segment 1 is transmitted, and turned off
when its acknowledgment (segment 2) arrives. Although its RTT is 1.061 seconds (from the
tcpdump output), examining the socket debug information shows that three of TCP’s clock ticks
occurred during this period, implying an RTT of 1500 ms.

The next segment timed is number 3. When segment 4 is transmitted 2.4 ms later, it cannot be timed,
since the timer for this connection is already in use. When segment 5 arrives, acknowledging the data
that was being timed, its RTT is calculated to be 1 tick (500 ms), even though we see that its RTT is
0.808 seconds from the tcpdump output.

The timer is started again when segment 6 is transmitted, and turned off when its acknowledgment
(segment 10) is received 1.015 seconds later. The measured RTT is 2 clock ticks. Segments 7 and 9
cannot be timed, since the timer is already being used. Also, when segment 8 is received (the ACK of
769), nothing is updated since the acknowledgment was not for bytes being timed.

Figure 21.3 shows the relationship in this example between the actual RTTs that we can determine
from the tcpdump output, and the counted clock ticks.

294

[IMAGE]
Figure 21.3 RTT measurements and clock ticks.

On the top we show the clock ticks, every 500 ms. On the bottom we show the times output by
tcpdump, and when the timer for the connection is turned on and off. We know that 3 ticks occur
between sending segment 1 and receiving segment 2, 1.061 seconds later, so we assume the first tick
occurs at time 0.03. (The first tick must be between 0.00 and 0.061.) The figure then shows how the
second measured RTT was counted as 1 tick, and the third as 2 ticks.

In this complete example, 128 segments were transmitted, and 18 RTT samples were collected. Figure
21.4 shows the measured RTT (taken from the tcpdump output) along with the RTO used by TCP for
the timeout (taken from the socket debug output). The x-axis starts at time 0 in Figure 21.2, when the
first data segment is transmitted, not when the first SYN is transmitted.

[IMAGE]
Figure 21.4 Measured RTT and TCP’s calculated RTO for example.

The first three data points for the measured RTT correspond to the 3 RTTs that we show in Figure
21.2. The gaps in the RTT samples around times 10, 14, and 21 are caused by retransmissions that took
place there (which we’ll show later in this chapter). Kam’s algorithm prevents us from updating our
estimators until another segment is transmitted and acknowledged. Also note that for this
implementation, TCP’s calculated RTO is always a multiple of 500 ms.

RTT Estimator Calculations
Let’s see how the RTT estimators (the smoothed RTT and the smoothed mean deviation) are
initialized and updated, and how each retransmission timeout is calculated.

The variables A and D are initialized to 0 and 3 seconds, respectively. The initial retransmission
timeout is calculated using the formula

RTO = A + 2D = 0 + 2x3 = 6 seconds

(The factor 2D is used only for this initial calculation. After this 4D is added to A to calculate RTO, as
shown earlier.) This is the RTO for the transmission of the initial SYN.

It turns out that this initial SYN is lost, and we time out and retransmit. Figure 21.5 shows the first
four lines from the tcpdump output file.

295

slip.1024 > vangogh. discard: S
1 0.0 35648001:35648001(0)
win 4096 <mss 256>
slip.1024 > vangogh.discard: S
5.802377
2 35648001:35648001(0)
(5.8024)
win 4096 <mss 256>
vangogh.discard > slip.1024: S
6.269395
3 1365512705:1365512705(0)
(0.4670)
ack 35648002 win 8192 <mss 512>
6.270796
4 slip. 1024 > vangogh.discard: . ack 1 win 4096
(0.0014)

Figure 21.5 Timeout and retransmission of initial SYN.

When the timeout occurs after 5.802 seconds, the current RTO is calculated as

RTO = A + 4D = 0 + 4x3 = 12 seconds
The exponential backoff is then applied to the RTO of 12. Since this is the first timeout we use a
multiplier of 2, giving the next timeout value as 24 seconds. The next timeout is calculated using a
multiplier of 4, giving a value of 48 seconds: 12 x 4. (These initial RTOs for the first SYN on a
connection, 6 seconds and then 24 seconds, are what we saw in Figure 4.5.)

The ACK arrives 467 ms after the retransmission. The values of A and D are not updated, because of
Karn’s algorithm dealing with the retransmission ambiguity. The next segment sent is the ACK on line
4, but it is not timed since it is only an ACK. (Only segments containing data are timed.)

When the first data segment is sent (segment 1 in Figure 21.2) the RTO is not changed, again owing to
Karn’s algorithm. The current value of 24 seconds is reused until an RTT measurement is made. This
means the RTO for time 0 in Figure 21.4 is really 24, but we didn’t plot that point.

When the ACK for the first data segment arrives (segment 2 in Figure 21.2), three clock ticks were
counted and our estimators are initialized as

A = M + 0.5 = 1.5 + 0.5 = 2
D = A/2 = 1
(The value 1.5 for M is for 3 clock ticks.) The previous initialization of A and D to 0 and 3 was for the
initial RTO calculation. This initialization is for the first calculation of the estimators using the first
RTT measurement M. The RTO is calculated as

RTO = A + 4D = 2 + 4x1 = 6 seconds
When the ACK for the second data segment arrives (segment 5 in Figure 21.2), 1 clock tick is counted
(0.5 seconds) and our estimators are updated as

Err = M - A = 0.5 - 2 = -1.5
A = A + gErr = 2 - 0.125x1.5 = 1.8125

296

D = D + h(|Err| - D) = 1 + 0.25x(1.5 - 1) = 1.125
RTO = A + 4D = 1.8125 + 4x1.125 = 6.3125
There are some subtleties in the fixed-point representations of Err, A, and D, and the fixed-point
calculations that are actually used (which we’ve shown in floating-point for simplicity). These
differences yield an RTO of 6 seconds (not 6.3125), which is what we plot in Figure 21.4 for time
1.871.

Slow Start
We described the slow start algorithm in Section 20.6. We can see it in action again in Figure 21.2.

Only one segment is initially transmitted on the connection, and its acknowledgment must be received
before another segment is transmitted. When segment 2 is received, two more segments are
transmitted.

21.5 Congestion Example
Now let’s look at the transmission of the data segments. Figure 21.6 is a plot of the starting sequence
number in a segment versus the time that the segment was sent. This provides a nice way to visualize
the data transmission. Normally the data points should move up and to the right, with the slope of the
points being the transfer rate. Retransmissions will appear as motion down and to the right.

At the beginning of Section 21.4 we said the total time for the transfer was about 45 seconds, but we
show only 35 seconds in this figure. These 35 seconds account for sending the data segments only.
The first data segment was not transmitted until 6.3 seconds after the first SYN was sent, because the
first SYN appears to have been lost and was retransmitted (Figure 21.5). Also, after the final data
segment and the FIN were sent (at time 34.1 in Figure 21.6) it took another 4.0 seconds to receive the
final 14 ACKs from the receiver, before the receiver’s FIN was received.

[IMAGE]
Figure 21.6 Sending of 32768 bytes of data from slip to vangogh.

We can immediately see the three retransmissions around times 10, 14, and 21 in Figure 21.6. At each
of these three points we can also see that only one segment is retransmitted, because only one dot dips
below the upward slope.

Let’s examine the first of these dips in detail (around the 10-second mark). From the tcpdump output
we can put together Figure 21.7.

[IMAGE]
Figure 21.7 Packet exchange for retransmission around the 10-second mark.

We have removed all the window advertisements from this figure, except for segment 72, which we
discuss below, slip always advertised a window of 4096, and vangogh advertised a window of
8192. The segments are numbered in this figure as a continuation of Figure 21.2, where the first data
segment across the connection was numbered 1. As in Figure 21.2, the segments are numbered
according to their send or receive order on the host slip, where tcpdump was being run. We have
also removed a few segments that have no relevance to the discussion (44, 47, and 49, all ACKs from
vangogh).

297

It appears that segment 45 got lost or arrived damaged-we can’t tell from this output. What we see on
the host slip is the acknowledgment for everything up through but not including byte 6657 (segment
58), followed by eight more ACKs of this same sequence number. It is the reception of segment 62,
the third of the duplicate ACKs, that forces the retransmission of the data starting at sequence number
6657 (segment 63). Indeed, Berkeley-derived implementations count the number of duplicate ACKs
received, and when the third one is received, assume that a segment has been lost and retransmit only
one segment, starting with that sequence number. This is Jacobson’s fast retransmit algorithm, which
is followed by his fast recovery algorithm. We discuss both algorithms in Section 21.7.

Notice that after the retransmission (segment 63), the sender continues normal data transmission
(segments 67, 69, and 71). TCP does not wait for the other end to acknowledge the retransmission.

Let’s examine what happens at the receiver. When normal data is received in sequence (segment 43),
the receiving TCP passes the 256 bytes of data to the user process. But the next segment received
(segment 46) is out of order: the starting sequence number of the data (6913) is not the next expected
sequence number (6657). TCP saves the 256 bytes of data and responds with an ACK of the highest
sequence number successfully received, plus one (6657). The next seven segments received by
vangogh (48, 50, 52, 54, 55, 57, and 59) are also out of order. The data is saved by the receiving
TCP, and duplicate ACKs are generated.

Currently there is no way for TCP to tell the other end that a segment is missing. Also, TCP cannot
acknowledge out-of-order data. All vangogh can do at this point is continue sending the ACKs of
6657.

When the missing data arrives (segment 63), the receiving TCP now has data bytes 6657-8960 in its
buffer, and passes these 2304 bytes to the user process. All 2304 bytes are acknowledged in segment
72. Also notice that this ACK advertises a window of 5888 (8192 - 2304), since the user process hasn’t
had a chance to read the 2304 bytes that are ready for it.

If we look in detail at the tcpdump output for the dips around times 14 and 21 in Figure 21.6, we see
that they too were caused by the receipt of three duplicate ACKs, indicating that a packet had been
lost. In each of these cases only a single packet was retransmitted.

We’ll continue this example in Section 21.8, after describing more about the congestion avoidance
algorithms.

21.6 Congestion Avoidance Algorithm
Slow start, which we described in Section 20.6, is the way to initiate data flow across a connection.
But at some point we’ll reach the limit of an intervening router, and packets can be dropped.
Congestion avoidance is a way to deal with lost packets. It is described in [Jacobson 1988].

The assumption of the algorithm is that packet loss caused by damage is very small (much less than
1%), therefore the loss of a packet signals congestion somewhere in the network between the source
and destination. There are two indications of packet loss: a timeout occurring and the receipt of
duplicate ACKs. (We saw the latter in Section 21.5. If we are using a timeout as an indication of
congestion, we can see the need for a good RTT algorithm, such as that described in Section 21.3.)

Congestion avoidance and slow start are independent algorithms with different objectives. But when
congestion occurs we want to slow down the transmission rate of packets into the network, and then
invoke slow start to get things going again. In practice they are implemented together.

298

Congestion avoidance and slow start require that two variables be maintained for each connection: a
congestion window, cwnd, and a slow start threshold size, ssthresh.The combined algorithm operates
as follows:

1. Initialization for a given connection sets cwnd to one segment and ssthresh to 65535 bytes.
2. The TCP output routine never sends more than the minimum of cwnd and the receiver’s
advertised window.

Congestion avoidance is flow control imposed by the sender, while the advertised window is flow
control imposed by the receiver. The former is based on the sender’s assessment of perceived
network congestion; the latter is related to the amount of available buffer space at the receiver for
this connection.
3. When congestion occurs (indicated by a timeout or the reception of duplicate ACKs), one-half of
the current window size (the minimum of cwnd and the receiver’s advertised window, but at least
two segments) is saved in ssthresh. Additionally, if the congestion is indicated by a timeout, cwnd
is set to one segment (i.e., slow start).
4. When new data is acknowledged by the other end, we increase cwnd, but the way it increases
depends on whether we’re performing slow start or congestion avoidance.

If cwnd is less than or equal to ssthresh, we’re doing slow start; otherwise we’re doing congestion
avoidance. Slow start continues until we’re halfway to where we were when congestion occurred
(since we recorded half of the window size that got us into trouble in step 2), and then congestion
avoidance takes over.

Slow start has cwnd start at one segment, and be incremented by one segment every time an ACK
is received. As mentioned in Section 20.6, this opens the window exponentially: send one
segment, then two, then four, and so on.

Congestion avoidance dictates that cwnd be incremented by 1/cwnd each time an ACK is
received. This is an additive increase, compared to slow start’s exponential increase. We want to
increase cwnd by at most one segment each round-trip time (regardless how many ACKs are
received in that RTT), whereas slow start will increment cwnd by the number of ACKs received
in a round-trip time.

All 4.3BSD releases and 4.4BSD incorrectly add a small fraction of the segment size (the segment size divided by 8)
during congestion avoidance. This is wrong and should not be emulated in future releases [Floyd 1994].
Nevertheless, we show this term in future calculations, to arrive at the same answer as the (incorrect)
implementation.

The 4.3BSD Tahoe release, described in [Leffler et al. 1989], performed slow start only if the other end was on a
different network. This was changed with the 4.3BSD Reno release so that slow start is always performed.

Figure 21.8 is a visual description of slow start and congestion avoidance. We show cwnd and ssthresh
in units of segments, but they’re really maintained in bytes.

[IMAGE]
Figure 21.8 Visualization of slow start and congestion avoidance.

In this figure we assume that congestion occurred when cwnd had a value of 32 segments. ssthresh is
then set to 16 segments and cwnd is set to 1 segment. One segment is then sent at time 0 and assuming
its ACK is returned at time 1, cwnd is incremented to 2 segments. Two segments are then sent and
assuming their ACKs return by time 2, cwndis incremented to 4 segments (once for each ACK). This
exponential increase continues until cwnd equals ssthresh, after 8 ACKs are received between times 3

299

and 4. From this point on the increase in cwnd is linear, with a maximum increase of one segment per
round-trip time.

As we can see in this figure, the term "slow start" is not completely correct. It is a slower transmission
of packets than what caused the congestion, but the rate of increase in the number of packets injected
into the network increases during slow start. The rate of increase doesn’t slow down until ssthresh is
reached, when congestion avoidance takes over.

21.7 Fast Retransmit and Fast Recovery Algorithms
Modifications to the congestion avoidance algorithm were proposed in 1990 [Jacobson 1990b]. We’ve
already seen these modifications in action in our congestion example (Section 21.5).

Before describing the change, realize that TCP is required to generate an immediate acknowledgment
(a duplicate ACK) when an out-of-order segment is received. This duplicate ACK should not be
delayed. The purpose of this duplicate ACK is to let the other end know that a segment was received
out of order, and to tell it what sequence number is expected.

Since we don’t know whether a duplicate ACK is caused by a lost segment or just a reordering of
segments, we wait for a small number of duplicate ACKs to be received. It is assumed that if there is
just a reordering of the segments, there will be only one or two duplicate ACKs before the reordered
segment is processed, which will then generate a new ACK. If three or more duplicate ACKs are
received in a row, it is a strong indication that a segment has been lost. (We saw this in Section 21.5.)
We then perform a retransmission of what appears to be the missing segment, without waiting for a
retransmission timer to expire. This is the fast retransmit algorithm. Next, congestion avoidance, but
not slow start is performed. This is the fast recovery algorithm.

In Figure 21.7 we saw that slow start was not performed after the three duplicate ACKs were received.
Instead the sender did the retransmission, followed by three more segments with new data (segments
67, 69, and 71), before the acknowledgment of the retransmission was received (segment 72).

The reason for not performing slow start in this case is that the receipt of the duplicate ACKs tells us
more than just a packet has been lost. Since the receiver can only generate the duplicate ACK when
another segment is received, that segment has left the network and is in the receiver’s buffer. That is,
there is still data flowing between the two ends, and we don’t want to reduce the flow abruptly by
going into slow start. This algorithms are usually implemented together as follows.

1. When the third duplicate ACK is received, set ssthresh to one-half the current congestion
window, cwnd.

Retransmit the missing segment. Set cwnd to ssthresh plus 3 times the segment size.
2. Each time another duplicate ACK arrives, increment cwnd by the segment size and transmit a
packet (if allowed by the new value of cwnd).
3. When the next ACK arrives that acknowledges new data, set cwnd to ssthresh (the value set in
step 1). This should be the ACK of the retransmission from step 1, one round-trip time after the
retransmission. Additionally, this ACK should acknowledge all the intermediate segments sent
between the lost packet and the receipt of the first duplicate ACK. This step is congestion
avoidance, since we’re slowing down to one-half the rate we were at when the packet was lost.

300

We’ll see what happens to the two variables cwnd and ssthresh in the calculations in the next section.

The fast retransmit algorithm first appeared in the 4.3BSD Tahoe release, but it was incorrectly followed by slow start. The
fast recovery algorithm appeared in the 4.3BSD Reno release.

21.8 Congestion Example (Continued)
Watching a connection using tcpdump and the socket debug option (which we described in Section
21.4) we can see the values of cwnd and ssthresh as each segment is transmitted. If the MSS is 256
bytes, the initial values of cwnd and ssthresh are 256 and 65535, respectively. Each time an ACK is
received we can see cwnd incremented by the MSS, taking on the values 512, 768, 1024, 1280, and so
on. Assuming congestion doesn’t occur, eventually the congestion window will exceed the receiver’s
advertised window, meaning the advertised window will limit the data flow.

A more interesting example is to see what happens when congestion occurs. We’ll use the same
example from Section 21.4. There were four occurrences of congestion while this example was being
run. There was a timeout on the transmission of the initial SYN to establish the connection (Figure
21.5), followed by three lost packets during the data transfer (Figure 21.6).

Figure 21.9 shows the values of the two variables cwnd and ssthresh when the initial SYN is
retransmitted, followed by the first seven data segments. (We showed the exchange of the initial data
segments and their ACKs in Figure 21.2.) We show the data bytes transmitted using the tcpdump
notation: 1:257(256) means bytes 1 through 256.

Segment# Action Variable
(Figure 21.2) Send Receive Comment cwnd ssthresh
initialize
SYN SYN, ACK 256 65535
timeout
SYN 256 512
retransmit
ACK - - -
-
1
2
3 1:257(256)
4 257:513(256)
ACK 257 slow start 512 512
5 513:769(256)
ACK 513 slow start 768 512
6 769:1025(256)
ACK 769 cong. avoid 885 512
7 1025:1281(256)
8 1281:1537(256)
9 1537:1793(256)
10 -
11
12

Figure 21.9 Example of congestion avoidance.

When the timeout of the SYN occurs, ssthresh is set to its minimum value (512 bytes. which is two
segments for this example), cwnd is set to one segment (256 bytes, which it was already at) to enter the
slow start phase.

301

When the SYN and ACK are received, nothing happens to the two variables, since new data is not
being acknowledged.

When the ACK 257 arrives, we are still in slow start since cwnd is less than or equal to ssthresh, so
cwnd in incremented by 256. The same thing happens when the ACK 512 arrives.

When the ACK 769 arrives we are no longer in slow start, but enter congestior avoidance. The new
value for cwnd is calculated as

cwnd <- cwnd + segsize x segsize/cwnd + segsize/8
This is the 1/cwnd increase that we mentioned earlier, taking into account that cwnd is really
maintained in bytes and not segments. For this example we calculate

cwnd <- 768 + 256 x 256 / 768 + 256 / 8
which equals 885 (using integer arithmetic). When the next ACK 1025 arrives we calculate

cwnd <- 885 + 256x256/885 + 256/8
which equals 991. (In these expressions we include the incorrect 256/8 term to match the values
calculated by the implementation, as we noted in Section 21.6)

This additive increase in cwnd continues until the first retransmission, around the 10-second mark in
Figure 21.6. Figure 21.10 is a plot of the same data as in Figure 21.6, with the value of cwnd added.

[IMAGE]
Figure 21.10 Value of cwnd and send sequence number while data is being transmitted.

The first six values for cwnd in this figure are the values we calculated for Figure 21.9. It is impossible
in this figure to tell the difference visually between the exponential increase during slow start and the
additive increase during congestion avoidance, because the slow start phase is so quick.

We need to explain what is happening at the three points where a retransmission occurs. Recall that
each of the retransmissions took place because three duplicate ACKs were received, indicating a
packet had been lost. This is the fast retransmit algorithm from Section 21.7. ssthresh is immediately
set to one-half the window size that was in effect when the retransmission took place, but cwnd is
allowed to keep increasing while the duplicate ACKs are received, since each duplicate ACK means
that a segment has left the network (the receiving TCP has buffered it, waiting for the missing hole in
the data to arrive). This is the fast recovery algorithm.

Figure 21.11 is similar to Figure 21.9, showing the values of cwnd and ssthresh. The segment numbers
in the first column correspond to Figure 21.7.

302

Segment# Action Variable
(Figure 21.7) Send Receive Comment cwnd ssthresh
58
59
60 ACK of new data
ACK 6657 2426 512
61 duplicate ACK #1
ACK 6657 2426 512
62 duplicate ACK #2
8705:8961(256) ACK 6657 2426 512
63 duplicate ACK #3
6657:6913(256) ACK 6657 1792 1024
64 retransmission
8961:9217(256) ACK 6657 2048 1024
65 duplicate ACK #4
9217:9473(256) ACK 6657 2304 1024
66 duplicate ACK #5
9473:9729(256) ACK 6657 2560 1024
67 duplicate ACK #6
. ACK 6657 2816 1024
68 duplicate ACK #7
ACK 6657 3072 1024
69 duplicate ACK #8
ACK 8961 1280 1024
70 ACK of new data
71
72

Figure 21.11 Example of congestion avoidance (continued).

The values for cwnd have been increasing continually, from the final value in Figure 21.9 for segment
12 (1089), to the first value in Figure 21.11 for segment 58 (2426). The value of ssthresh has remained
the same (512), since there have been no retransmissions in this period.

When the first two duplicate ACKs arrive (segments 60 and 61) they are counted, and cwnd is left
alone. (This is the flat portion of Figure 21.10 preceding the retransmission.) When the third one
arrives, however, ssthresh is set to one-half cwnd (rounded down to the next multiple of the segment
size), cwnd is set to ssthresh plus the number of duplicate ACKs times the segment size (i.e., 1024
plus 3 times 256). The retransmission is then sent.

Five more duplicate ACKs arrive (segments 64-66, 68, and 70) and cwnd is incremented by the
segment size each time. Finally a new ACK arrives (segment 72) and cwnd is set to ssthresh (1024)
and the normal congestion avoidance takes over. Since cwnd is less than or equal to ssthresh (they are
equal), the segment size is added to cwnd, giving a value of 1280. When the next new ACK is received
(which isn’t shown in Figure 21.11), cwnd is greater than ssthresh, so cwnd is set to 1363.

During the fast retransmit and fast recovery phase, we transmit new data after receiving the duplicate
ACKs in segments 66, 68, and 70, but not after receiving the duplicate ACKs in segments 64 and 65.
The reason is the value of cwnd, versus the number of unacknowledged bytes of data. When segment
64 is received, cwnd equals 2048, but we have 2304 unacknowledged bytes (nine segments: 46, 48, 50,
52, 54, 55, 57, 59, and 63). We can’t send anything. When segment 65 arrives, cwnd equals 2304, so
we still can’t send anything. But when segment 66 arrives, cwnd equals 2560, so we can send a new
data segment. Similarly when segment 68 arrives, cwnd equals 2816, which is greater than the 2560
bytes of unacknowledged data, so we can send another new data segment. The same scenario happens
when segment 70 is received.

W^hen the next retransmission takes place at time 14.3 in Figure 21.10, it is also triggered by the
reception of three duplicate ACKs, so we see the same increase in cwnd as one more duplicate ACK
arrives, followed by a decrease to 1024.

303

The retransmission at time 21.1 in Figure 21.10 is also triggered by duplicate ACKs. We receive three
more duplicates after the retransmission, so we see three additional increases in cwnd, followed by a
decrease to 1280. For the remainder of the transfer cwnd increases linearly to a final value of 3615.

21.9 Per-Route Metrics
Newer TCP implementations maintain many of the metrics that we’ve described in this chapter in the
routing table entry. When a TCP connection is closed, if enough data was sent to obtain meaningful
statistics, and if the routing table entry for the destination is not a default route, the following
information is saved in the routing table entry, for the next use of the entry: the smoothed RTT, the
smoothed mean deviation, and the slow start threshold. The quantity "enough data" is 16 windows of
data. This gives 16 RTT samples, which allows the smoothed RTT filter to converge within 5% of the
correct value.

Additionally, the route(8) command can be used by the administrator to set the metrics for a given
route: the three values mentioned in the preceding paragraph, along with the MTU, the outbound
bandwidth-delay product (Section 20.7), and the inbound bandwidth-delay product.

When a new TCP connection is established, either actively or passively, if the routing table entry being
used for the connection has values for these metrics, the corresponding variable is initialized from the
metrics.

21.10 ICMP Errors
Let’s see how TCP handles ICMP errors that are returned for a given connection. The most common
ICMP errors that TCP can encounter are source quench, host unreach-able, and network unreachable.
Current Berkeley-based implementations handle these ICMP errors as follows:

A received source quench causes the congestion window, cwnd, to be set to one segment to
initiate slow start, but the slow start threshold, ssthresh, is not changed, so the window will open
until it’s either open all the way (limited by the window size and round-trip time) or until
congestion occurs.
A received host unreachable or network unreachable is effectively ignored, since these two errors
are considered transient. It could be that an intermediate router has gone down and it can take the
routing protocols a few minutes to stabilize on an alternative route. During this period either of
these two ICMP errors can occur, but they must not abort the connection. Instead, TCP keeps
trying to send the data that caused the error, although it may eventually time out. (Recall in
Figure 21.1 that TCP did not give up for 9 minutes.)

Current Berkeley-based implementations record that the ICMP error occurred, and if the
connection times out, the ICMP error is translated into a more relevant error code than
"connection timed out."

Earlier BSD implementations incorrectly aborted a connection whenever an ICMP host unreachable or network
unreachable was received.

An Example
We can see how an ICMP host unreachable is handled by taking our dialup SLIP link down during the
middle of a connection. We establish a connection from the host slip to the host aix. (From the
figure on the inside front cover we see that this connection goes through our dialup SLIP link.) After

304

establishing the connection and transferring some data, the dialup SLIP link between the routers sun
and netb is taken down. This causes the default routing table entry on sun (which we showed in
Section 9.2) to be removed. We expect sun to then respond to IP datagrams destined for the 140.252.1
Ethernet with an ICMP host unreachable. We want to see how TCP handles these ICMP errors.

Here is the interactive session on the host slip:

slip % sock aix echo run our sock program
test line type this line
test line and it’s echoed
SLIP link is brought down here
another line then type this line and watch retransmissions
SLIP link is reestablished here
another line and the line and its echo are exchanged
line number 3
line number 3
the last line
SLIP link is brought down here, and not
reestablished
read error: No route to
TCP finally gives up
host

Figure 21.12 shows the corresponding tcpdump output, captured on the router bsdi. (We have
removed the connection establishment and all the window advertisements.) We connect to the echo
server on the host aix and type "test line" (line 1). It is echoed (line 2) and the echo is acknowledged
(line 3). We then take down the SLIP link.

slip.1035 > aix. echo: P 1:11(10) ack
1 0.0
1
aix.echo > slip.1035: P 1:11(10) ack
2 0.212271 ( 0.2123)
11
3 0.310685 ( 0.0984) slip.1035 > aix.echo: . ack 11
SLIP link brought down here
slip.1035 > aix.echo: P 11:24(13) ack
4 174.758100 (174.4474)
11
5 174.759017 ( 0.0009) sun > slip: icmp: host aix unreachable
6 177.150439 ( 2.3914)
11

305

8 182.150200 ( 4.9989)
11
10 192.149671 ( 9.9985)
11
12 212.148783 ( 19.9982)
11
SLIP link brought up here
14 252.146774 ( 39.9970)
11
15 252.439257 ( 0.2925)
24
16 252.505331 ( 0.0661) slip.1035 > aix.echo: . ack 24
17 261.977246 ( 9.4719)
24
18 262.158758 ( 0.1815)
38
19 262.305086 ( 0.1463) slip.1035 > aix.echo: . ack 38
SLIP link brought down here
20 458.155330 (195.8502)
38
22 461.136904 (2.9807)
38
23 461.137826 (0.0009) sun > slip: icmp: host aix unreachable
24 467.136461 (5.9986)
38
25 467.137385 (0.0009) sun > slip: icmp: host aix unreachable
26 479.135811 ( 11.9984)
38

306

28 503.134816 ( 23.9982)
38
29 503.135740 ( 0.0009) sun > slip: icmp: host aix unreachabie
14 lines of output deleted here
1000.219573 ( slip.1035 > aix.echo: P 38:52(14) ack
44
64.0959) 38
1064.201281 ( slip.1035 > aix.echo: R 52:52(0) ack
46
63.9808) 38

Figure 21.12 TCP handling of received ICMP host unreachable error.

We type "another line" (line 3) and expect to see TCP time out and retransmit the message. Indeed,
this line is sent six times before a reply is received. Lines 4-13 show the first transmission and the next
four retransmissions, each of which generates an ICMP host unreachable from the router sun. "This is
what we expect: the IP datagrams go from slip to the router bsdi (which has a default route that
points to sun), and then to sun, where the broken link is detected.

While these retransmissions are taking place, the SLIP link is brought back up, and the retransmission
on line 14 gets delivered. Line 15 is the echo from aix, and line 16 is the acknowledgment of the
echo.

This shows that TCP ignores the ICMP host unreachable errors and keeps retransmitting. We can also
see the expected exponential backoff in each retransmission timeout: the first appears to be 2.5
seconds, which is then multiplied by 2 (giving 5 seconds), then 4 (10 seconds), then 8 (20 seconds),
then 16 (40 seconds).

We then type the third line of input ("line number 3") and see it sent on line 17, echoed on line 18, and
the echo acknowledged on line 19.

We now want to see what happens when TCP retransmits and gives up, after receiving the ICMP host
unreachable, so we take down the SLIP link again. After taking it down we type "the last line" and see
it transmitted 13 times before TCP gives up. (We have deleted lines 30-43 from the output. They are
additional retransmissions.)

The thing we notice, however, is the error message printed by our sock program when it finally gives
up: "No route to host." This corresponds to the Unix error associated with the ICMP host unreachable
error (Figure 6.12). This shows that TCP saves the ICMP error that it receives on the connection, and
when it finally gives up, it prints that error, instead of "Connection timed out."

Finally, notice the different retransmission intervals in lines 22-46, compared to lines 6-14. It appears
that TCP updated its estimators when the third line we typed was sent and acknowledged without any
retransmissions in lines 17-19. The initial retransmission timeout is now 3 seconds, giving successive
values of 6, 12, 24, 48, and then the upper limit of 64.

307

21.11 Repacketization
When TCP times out and retransmits, it does not have to retransmit the identical segment again.
Instead, TCP is allowed to perform repacketization, sending a bigger segment, which can increase
performance. (Naturally, this bigger segment cannot exceed the MSS announced by the other
receiver.) This is allowed in the protocol because TCP identifies the data being sent and acknowledged
by its byte number, not its segment number.

We can easily see this in action. We use our sock program to connect to the discard server and type
one line. We then disconnect the Ethernet cable and type a second line. While this second line is being
retransmitted, we type a third line. We expect the next retransmission to contain both the second and
third lines.

bsdi % sock svr4 discard
hello there first line gets sent OK
then we disconnect the Ethernet cable
line number 2 this line gets retransmitted
and 3 type this line before second line sent OK
then reconnect Ethernet cable

Figure 21.13 shows the tcpdump output. (We have removed the connection establishment, the
connection termination, and all the window advertisements.)

308

bsdi. 1032 > svr4.discard: P 1:13(12) ack
1 0.0
1
2 0.140489 ( 0.1405) svr4.discard > bsdi.1032: . ack 13
Ethernet cable disconnected here
26.407696 bsdi.1032 > svr4.discard: P 13:27(14) ack
3
(26.2672) 1
27.639390 ( bsdi.1032 > svr4.discard: P 13:27(14) ack
4
1.2317) 1
5
3.0001) 1
third line typed here
6
6.0002) 1
7
(12.0005) 1
8
(24.0006) 1
Ethernet cable reconnected here
72.719091 (
9 svr4.discard > bsdi.1032: . ack 33
0.0783)

Figure 21.13 Repacketization of data by TCP.

Lines 1 and 2 show the first line ("hello there") being sent and its acknowledgment. We then
disconnect the Ethernet cable and type "line number 2" (14 bytes, including the newline). These bytes
are transmitted on line 3, and then retransmitted on lines 4 and 5. Before the retransmission on line 6
we type "and 3" (6 bytes, including the newline) and see this retransmission contain 20 bytes: both
lines that we typed. When the acknowledgment arrives on line 9, it is for all 20 bytes.

21.12 Summary
This chapter has provided a detailed look at TCP’s timeout and retransmission strategy. Our first
example was a lost SYN to establish a connection and we saw how an exponential backoff is applied
to successive retransmission timeout values.

TCP calculates the round-trip time and then uses these measurements to keep track of a smoothed RTT
estimator and a smoothed mean deviation estimator. These two estimators are then used to calculate
the next retransmission timeout value. Many implementations only measure a single RTT per window.
Karn’s algorithm removes the retransmission ambiguity problem by preventing us from measuring the
RTT when a packet is lost.

309

Our detailed example, which included three lost packets, let us see many of TCP’s algorithms in
action: slow start, congestion avoidance, fast retransmit, and fast recovery. We were also able to hand
calculate TCP RTT estimators along with the congestion window and slow-start threshold, and verify
the values with the actual values from the trace output.

We finished the chapter by looking at the effect various ICMP errors have on a TCP connection and
how TCP is allowed to repacketize its data. We saw how the "soft" ICMP errors don’t cause a
connection to be terminated, but are remembered so that if the connection terminates abnormally, the
soft error can be reported.

Exercises
21.1 In Figure 21.5 the first timeout was calculated as 6 seconds and the next as 12 seconds. If the
ACK for the initial SYN had not arrived after the 12-second timeout expired, when would the next
timeout occur?

21.2 In the discussion following Figure 21.5 we said that the timeout intervals are calculated as 6, 24,
and then 48 seconds, as we saw in Figure 4.5. But if we watch a TCP connection to a nonexistent host
from an SVR4 system, the timeout intervals are 6, 12, 24, and 48 seconds. What’s going on?

21.3 Compare the performance of TCP’s sliding window versus TFTP’s stop-and-wait protocol as
follows. In this chapter we transferred 32768 bytes in about 35 seconds (Figure 21.6) across a link
with an RTT that averaged around 1.5 seconds (Figure 21.4). Calculate how long TFTP would take for
the same transfer.

21.4 In Section 21.7 we said that the receipt of a duplicate ACK is caused by a segment being lost or
reordered. In Section 21.5 we saw the generation of duplicate ACKs caused by a lost segment. Draw a
picture showing that a reordering of segments also generates duplicate ACKs.

21.5 There is a noticeable blip in Figure 21.6 between times 28.8 and 29.8. Is this a retransmission?

21.6 In Section 21.6 we said that the 4.3BSD Tahoe release only performed slow start if the
destination was on a different network. How do you think "different network" was determined? (Hint:
Look at Appendix E.)

21.7 In Section 20.2 we said that TCP normally ACKs every other segment. But in Figure 21.2 we see
the receiver ACK every segment. Why?

21.8 Are per-route metrics really useful, given the prevalence of default routes?

310

TCP Persist Timer
22.1 Introduction
We’ve seen that TCP has the receiver perform flow control by specifying the amount of data it is
willing to accept from the sender: the window size. What happens when the window size goes to 0?
This effectively stops the sender from transmitting data, until the window becomes nonzero.

We saw this scenario in Figure 20.3. When the sender received segment 9, opening the window that
was shut down by segment 8, it immediately started sending data. TCP must handle the case of this
acknowledgment that opens the window (segment 9) being lost. Acknowledgments are not reliably
transmitted-that is, TCP does not ACK acknowledgments, it only ACKs segments containing data.

If an acknowledgment is lost, we could end up with both sides waiting for the other: the receiver
waiting to receive data (since it provided the sender with a nonzero window) and the sender waiting to
receive the window update allowing it to send. To prevent this form of deadlock from occurring the
sender uses a persist timer that causes it to query the receiver periodically, to find out if the window
has been increased. These segments from the sender are called window probes. In this chapter we’ll
examine window probes and the persist timer. We’ll also examine the silly window syndrome, which
is tied to the persist timer.

22.2 An Example
To see the persist timer in action we’ll start a receiver process that listens for a connection request
from a client, accepts the connection request, and then goes to sleep for a long time before reading
from the network.

Our sock program lets us specify a pause option -P that sleeps between the server accepting the
connection request and performing the first read. We’ll invoke the server as:

svr4 % sock -i -s -P100000 5555

This has the server sleep for 100,000 seconds (27.8 hours) before reading from the network. The client
is run on the host bsdi and performs 1024-byte writes to port 5555 on the server. Figure 22.1 shows
the tcpdump output. (We have removed the connection establishment from the output.)

bsdi.1027 > svr4.5555: P 1:1025(1024) ack
1 0.0
1 win 4096
0.191961 (
2 svr4.5555 > bsdi.1027: . ack 1025 win 4096
0.1920)
0.196950 ( bsdi.1027 > svr4.5555: . 1025 :2049(1024)
3
0.0050) ack 1 win 4096
0.200340 ( bsdi.1027 > svr4.5555: . 2049:3073(1024)
4
0.0034) ack 1 win 4096
0.207506 (
5 svr4.5555 > bsdi.1027: . ack 3073 win 4096
0.0072)

311

0.212676 ( bsdi.1027 > svr4.5555: . 3073 :4097(1024)
6
0.0052) ack 1 win 4096
0.216113 ( bsdi.1027 > svr4.5555: P 4097 :5121(1024)
7
0.0034) ack 1 win 4096
0.219997 ( bsdi.1027 > svr4.5555: P 5121 :6145(1024)
8
0.0039) ack 1 win 4096
0.227882 (
9 svr4.5555 > bsdi.1027: . ack 5121 win 4096
0.0079)
0.233012 ( bsdi.1027 > svr4.5555: P 6145 :7169(1024)
10
0.0051) ack 1 win 4096
0.237014 ( bsdi.1027 > svr4.5555: P 7169 :8193(1024)
11
0.0040) ack 1 win 4096
0.240961 ( bsdi.1027 > svr4.5555: P 8193 :9217(1024)
12
0.0039) ack 1 win 4096
0.402143 (
13 svr4.5555 > bsdi.1027: . ack 9217 win 0
0.1612)
5.351561 ( bsdi.1027 > svr4.5555: . 9217 : 9218(1)
14
4.9494) ack 1 win 4096
5.355571 (
15 svr4.5555 > bsdi.1027: . ack 9217 win 0
0.0040)
10.351714 ( bsdi.1027 > svr4.5555: . 9217 :9218 (1)
16
4.9961) ack 1 win 4096
10.355670 (
17 svr4.5555 > bsdi.1027: . ack 9217 win 0
0.0040)
16.351881 ( bsdi.1027 > svr4.5555: . 9217 : 9218(1)
18
5.9962) ack 1 win 4096
16.355849 (
19 svr4.5555 > bsdi.1027: . ack 9217 win 0
0.0040)
28.352213 bsdi.1027 > svr4.5555: . 9217 :9218 (1)
20
(11.9964) ack 1 win 4096
28.356178 (
21 svr4.5555 > bsdi.1027: . ack 9217 win 0
0.0040)
52.352874 bsdi.1027 > svr4.5555: . 9217 : 9218(1)
22
(23.9967) ack 1 win 4096
52.356839 (
23 svr4.5555 > bsdi.1027: . ack 9217 win 0
0.0040)
100.354224 bsdi.1027 > svr4.5555: . 9217 : 9218(1)
24
(47.9974) ack 1 win 4096

312

100.358207 (
25 svr4.5555 > bsdi.1027: . ack 9217 win 0
0.0040)
160.355914 bsdi.1027 > svr4.5555: . 9217 :9218 (1)
26
(59.9977) ack 1 win 4096
160.359835 (
27 svr4.5555 > bsdi.1027: . ack 9217 win 0
0.0039)
220.357575 bsdi.1027 > svr4.5555: . 9217:9218(1) ack
28
(59.9977) 1 win 4096
220.361668 (
29 svr4.5555 > bsdi.1027: . ack 9217 win 0
0.0041)
280.359254 bsdi.1027 > svr4.5555: . 9217 : 9218(1)
30
(59.9976) ack 1 win 4096
280.363315 (
31 svr4.5555 > bsdi.1027: . ack 9217 win 0
0.0041)

Figure 22.1 Example of persist timer probing a zero-sized window.

Segments 1-13 shows the normal data transfer from the client to the server, filling up the window with
9216 bytes of data. The server advertises a window of 4096, and has a default socket buffer size of
4096, but really accepts a total of 9216 bytes. This is some form of interaction between the TCP/IP
code and the streams subsystem in SVR4.

In segment 13 the server acknowledges the previous four data segments, but advertises a window of 0,
stopping the client from transmitting any more data. This causes the client to set its persist timer. If the
client doesn’t receive a window update when the timer expires, it probes the empty window, to see if a
window update has been lost. Since our server process is asleep, the 9216 bytes of data are buffered by
TCP, waiting for the application to issue a read.

Notice the spacing of the window probes by the client. The first (segment 14) is 4.949 seconds after
receiving the zero-sized window. The next (segment 16) is 4.996 seconds later. The spacing is then
about 6, 12, 24,48, and 60 seconds after the previous.

Why are the spacings always a fraction of a second less than 5,6,12, 24, 48, and 60? These probes are
triggered by TCP’s 500-rns timer expiring. When the timer expires, the window probe is sent, and a
reply is received about 4 ms later. The receipt of the reply causes the timer to be restarted, but the time
until the next clock tick is about 500 minus 4 ms.

The normal TCP exponential backoff is used when calculating the persist timer. The first timeout is
calculated as 1.5 seconds for a typical LAN connection. This is multiplied by 2 for a second timeout
value of 3 seconds. A multiplier of 4 gives the next value of 6, a multiplier of 8 gives a value of 12,
and so on. But the persist timer is always bounded between 5 and 60 seconds, which accounts for the
values we see in Figure 22.1.

The window probes contain 1 byte of data (sequence number 9217). TCP is always allowed to send 1
byte of data beyond the end of a closed window. Notice, however, that the acknowledgments returned
with the window size of 0 do not ACK this byte. (They ACK the receipt of all bytes through and
including byte number 9216.) Therefore this byte keeps being retransmitted.

313

The characteristic of the persist state that is different from the retransmission timeout in Chapter 21 is
that TCP never gives up sending window probes. These window probes continue to be sent at
60-second intervals until the window opens up or either of the applications using the connection is
terminated.

22.3 Silly Window Syndrome
Window-based flow control schemes, such as the one used by TCP, can fall victim to a condition
known as the silly window syndrome (SWS). When it occurs, small amounts of data are exchanged
across the connection, instead of full-sized segments [dark 1982].

It can be caused by either end: the receiver can advertise small windows (instead of waiting until a
larger window could be advertised) and the sender can transmit small amounts of data (instead of
waiting for additional data, to send a larger segment). Correct avoidance of the silly window syndrome
is performed on both ends.

1. The receiver must not advertise small windows. The normal algorithm is for the receiver not to
advertise a larger window than it is currently advertising (which can be 0) until the window can
be increased by either one full-sized segment (i.e the MSS being received) or by one-half the
receiver’s buffer space, whichever is smaller.
2. Sender avoidance of the silly window syndrome is done by not transmitting unless one of the
following conditions is true: (a) a full-sized segment can be sent, (b) we can send at least one-half
of the maximum sized window that the other end has ever advertised, or (c) we can send
everything we have and either we are not expecting an ACK (i.e., we have no outstanding
unacknowledged data) or the Nagle algorithm is disabled for this connection (Section 19.4).

Condition (b) deals with hosts that always advertise tiny windows, perhaps smaller than the
segment size. Condition (c) prevents us from sending small segments when we have
unacknowledged data that is waiting to be ACKed and the Nagle algorithm is enabled. If the
application is doing small writes (e.g., smaller than the segment size), it is condition (c) that
avoids the silly window syndrome.

These three conditions also let us answer the question: if the Nagle algorithm prevents us from
sending small segments while there is outstanding unacknowledged data, how small is small?
From condition (a) we see that "small" means the number of bytes is less than the segment size.
Condition (b) only comes into play with older, primitive hosts.

Condition (b) in step 2 requires that the sender keep track of the maximum window size advertised by
the other end. This is an attempt by the sender to guess the size of the other end’s receive buffer.
Although the size of the receiver buffer could decrease while the connection is established, in practice
this is rare.

An Example
We’ll now go through a detailed example to see the silly window syndrome avoidance in action, which
also involves the persist timer. We’ll use our sock program with the sending host, sun, doing six
1024-byte writes to the network:

sun % sock -i -n6 bsdi 7777

314

But we’ll put some pauses in the receiving process on the host bsdi, pausing 4 seconds before doing
the first read, and then pausing 2 seconds between successive reads. Additionally, the receiver issues
256-byte reads;

bsdi % sock -i -s -P4 -p2 -r256 7777

The reason for the initial pause is to let the receiver’s buffer fill, forcing it to stop the transmitter.
Since the receiver then performs small reads from the network, we expect to see the receiver perform
silly window syndrome avoidance.

Figure 22.2 is the time line for the transfer of the 6144 bytes of data. (We have deleted the connection
establishment.)

We also need to track what happens with the application that’s reading the data at each point in time,
along with the number of bytes currently in the receive buffer, and the number of bytes of available
space in the receive buffer. Figure 22.3 shows everything that’s happening.

[IMAGE]
Figure 22.2 Time line showing receiver avoidance of silly window syndrome.

Segment#
Time Action Receiver buffer
(Figure 22.2)
Send TCP Receive TCP Application data available
0.000 1 1:1025(1024) 1024 3072
0.002 2 1025:2049(1024) 2048 2048
0.003 3 2049:3073(1024) 3072 1024
0.005 4 3073:4097(1024) 4096 0
0.170 5 ACK 4097, win 0
3.99 read 256 3840 256
5.151 6 4097:4098(1) 3841 255
5.17 7 ACK 4098, win 0
5.99 read 256 3585 511
7.99 read 256 3329 767
9.99 read 256 3073 1023
10.151 8 4098:4099(1) 3074 1022
10.170 9 ACK 4099, win 0
11.99 read 256 2818 1278
13.99 read 256 2562 1534
15.151 10 4099:4100(1) 2563 1533

315

15.170 11 ACK 4100, win 1533
15.172 12 4100:5124(1024) 3587 509
15.370 13 ACK 5124, win 509
15.99 read 256 3331 765
17.99 read 256 3075 1021
19.99 read 256 2819 1277
20.151 14 5124:5633(509) 3328 768
20.170 15 ACK 5633, win 0
21.99 read 256 3072 1024
23.99 read 256 2816 1280
25.151 16 5633:5634(1) 2817 1279
25.170 17 ACK 5634, win 1279
25.171 18 5634:6145(511) 3328 768
25.174 19 ACK 6146, win 767
25.99 read 256 3072 1024
27.99 read 256 2816 1280
29.99 read 256 2560 1536
31.99 read 256 2304 1792
33.99 read 256 2048 2048
35.99 read 256 1792 2304
37.99 read 256 1536 2560
39.99 read 256 1280 2816
39.99 20 ACK 6146, win 2816
41.99 read 256 1024 3072
43.99 read 256 768 3328
45.99 read 256 512 3584
47.99 read 256 256 3840
49.99 read 256 0 4096
51.99 read 256 (EOF) 0 4096
51.991 21 ACK 6146, win 4096

316

51.992 22 ACK2

Figure 22.3 Sequence of events for receiver avoidance of the silly window syndrome.

In Figure 22.3 the first column is the relative point in time for each action. Those times with three
digits to the right of the decimal point are taken from the tcpdump output (Figure 22.2). Those times
with 99 to the right of the decimal point are the assumed times of the action on the receiving host.
(Having these relative times on the receiver contain 99 for the hundredths of a second correlates them
with segments 20 and 22 in Figure 22.2, the only two events on the receiver that we can see with
tcpdump that are triggered by a timeout on the receiving host. All the other packets that we see from
bsdi are triggered by the reception of a segment from the sender. It also makes sense, because this
would place the initial 4-second pause just before time 0 when the sender transmits the first data
segment. This is about when the receiver would get control, after receiving the ACK of its SYN in the
connection establishment.)

The amount of data in the receiver’s buffer increases when it receives data from the sender, and
decreases as the application reads data from the buffer. What we want to follow are the window
advertisements sent by the receiver to the sender, and what those window advertisements are. This lets
us see how the silly window syndrome is avoided by the receiver.

The first four data segments and the corresponding ACK (segments 1-5) show the sender filling the
receiver’s buffer. At that point the sender is stopped but it still has more data to send. It sets its persist
timer for its minimum value of 5 seconds.

When the persist timer expires, 1 byte of data is sent (segment 6). The receiving application has read
256 bytes from the receive buffer (at time 3.99), so the byte is accepted and acknowledged (segment
7). But the advertised window is still 0, since the receiver does not have room for either one full-sized
segment or one-half of its buffer. This is silly window avoidance by the receiver.

The sender’s persist timer is reset and goes off again 5 seconds later (at time 10.151). One byte is
again sent and acknowledged (segments 8 and 9). Again the amount of room in the receiver’s buffer
(1022 bytes) forces it to advertise a window of 0.

When the sender’s persist timer expires next, at time 15.151, another byte is sent and acknowledged
(segments 10 and 11). This time the receiver has 1533 bytes available in its buffer, so a nonzero
window is advertised. The sender immediately takes advantage of the window and sends 1024 bytes
(segment 12). The acknowledgment of these 1024 bytes (segment 13) advertises a window of 509
bytes. This appears to contradict what we’ve seen earlier with small window advertisements.

What’s happening here is that segment 11 advertised a window of 1533 bytes but the sender only
transmitted 1024 bytes. If the acknowledgment in segment 13 advertised a window of 0, it would
violate the TCP principle that a window cannot shrink by moving the right edge of the window to the
left (Section 20.3). That’s why the small window of 509 bytes must be advertised.

Next we see that the sender does not immediately transmit into this small window. This is silly
window avoidance by the sender. Instead it waits for another persist timer to expire at time 20.151,
when it sends 509 bytes. Even though it ends up sending this small segment with 509 bytes of data, it
waits 5 seconds before doing so, to see if an ACK arrives that opens up the window more. These 509
bytes of data leave only 768 bytes of available space in the receive buffer, so the acknowledgment
(segment 15) advertises a window of 0.

317

The persist timer goes off again at time 25.151, and the sender transmits 1 byte. The receive buffer
then has 1279 bytes of space, which is the window advertised in segment 17.

The sender has only 511 additional bytes of data to transmit, which it sends immediately upon
receiving the window advertisement of 1279 (segment 18). This segment also contains the FIN flag.
The receiver acknowledges the data and the FIN, advertising a window of 767. (See Exercise 22.2.)

Since the sending application issues a close after performing its six 1024-byte writes, the sender’s end
of the connection goes from the ESTABLISHED state to the FIN_WAIT_1 state, to the FIN_WAIT_2
state (Figure 18.12). It sits in this state until receiving a FIN from the other end. There is no timer in
this state (recall our discussion at the end of Section 18.6), since the FIN that it sent in segment 18 was
acknowledged in segment 19. This is why we see no further transmissions by the sender until it
receives the FIN (segment 21).

The receiving application continues reading 256 bytes of data every 2 seconds from the receive buffer.
Why is the ACK sent at time 39.99 (segment 20)? The amount of room in the receive buffer has gone
from its last advertised value of 767 (segment 19) to 2816 when the application reads at time 39.99.
This equals 2049 bytes of additional space in the receive buffer. Recalling the first rule at the start of
this section, the receiver now sends a window update because the amount of room has increased by
one-half the room in the receive buffer. This implies that the receiving TCP checks whether to send a
window update every time the application reads data from TCP’s receive buffer.

The final application read occurs at time 51.99 and the application receives an end-of-file notification,
since the buffer is empty. This causes the final two segments (21 and 22), which complete the
termination of the connection.

22.4 Summary
TCP’s persist timer is set by one end of a connection when it has data to send, but has been stopped
because the other end has advertised a zero-sized window. The sender keeps probing the closed
window using a retransmission interval similar to what we saw in Chapter 21. This probing of the
closed window continues indefinitely.

When we ran an example to see the persist timer we also encountered TCP’s avoidance of the silly
window syndrome. This is to prevent TCP from advertising small windows or from sending small
segments. In our example we saw avoidance of the silly window syndrome by both the sender and the
receiver.

Exercises
22.1 In Figure 22.3 notice the times of all the acknowledgments (segments 5, 7, 9, 11, 13, 15, and 17):
0.170, 5.170, 10.170, 15.170, 15.370, 20.170, and 25.170. Also notice the time differences between
receiving the data and sending the ACK: 164.5, 18.5, 18.7, 18.8, 198.3, 18.5, and 19.1 ms. Explain
what’s probably going on.

22.2 In Figure 22.3 at time 25.174 a window of 767 is advertised, but 768 bytes are available in the
receive buffer. Why the difference of 1 byte?

318

TCP Keepalive Timer
23.1 Introduction
Many newcomers to TCP/IP are surprised to learn that no data whatsoever flows across an idle TCP
connection. That is, if neither process at the ends of a TCP connection is sending data to the other,
nothing is exchanged between the two TCP modules. There is no polling, for example, as you might
find with other networking protocols. This means we can start a client process that establishes a TCP
connection with a server, and walk away for hours, days, weeks or months, and the connection remains
up. Intermediate routers can crash and reboot, phone lines may go down and back up, but as long as
neither host at the ends of the connection reboots, the connection remains established.

This assumes that neither application-the client or server-has application-level timers to detect
inactivity, causing either application to terminate. Recall at the end of Section 10.7 that BGP sends an
application probe to the other end every 30 seconds. This is an application timer that is independent of
the TCP keepalive timer.

There are times, however, when a server wants to know if the client’s host has either crashed and is
down, or crashed and rebooted. The keepalive timer, a feature of many implementations, provides this
capability.

Keepalives are not part of the TCP specification. The Host Requirements RFC provides three reasons not to use them: (1)
they can cause perfectly good connections to be dropped during transient failures, (2) they consume unnecessary
bandwidth, and (3) they cost money on an internet that charges by the packet. Nevertheless, many implementations provide
the keep-alive timer.

The keepalive timer is a controversial feature. Many feel that this polling of the other end has no place
in TCP and should be done by the application, if desired. This is one of the religious issues, because of
the fervor expressed by some on the topic.

The keepalive option can cause an otherwise good connection between two processes to be terminated
because of a temporary loss of connectivity in the network joining the two end systems. For example,
if the keepalive probes are sent during the time that an intermediate router has crashed and is
rebooting, TCP will think that the client’s host has crashed, which is not what has happened.

The keepalive feature is intended for server applications that might tie up resources on behalf of a
client, and want to know if the client host crashes. Many versions of the Telnet server and Rlogin
server enable the keepalive option by default.

A common example showing the need for the keepalive feature nowadays is when personal computer
users use TCP/IP to login to a host using Telnet. If they just power off the computer at the end of the
day, without logging off, they leave a half-open connection. In Figure 18.16 we showed how sending
data across a half-open connection caused a reset to be returned, but that was from the client end,
where the client was sending the data. If the client disappears, leaving the half-open connection on the
server’s end, and the server is waiting for some data from the client, the server will wait forever. The
keepalive feature is intended to detect these half-open connections from the server side.

319

23.2 Description
In this description we’ll call the end that enables the keepalive option the server, and the other end the
client. There is nothing to stop a client from setting this option, but normally it’s set by servers. It can
also be set by both ends of a connection, if it’s important for each end to know if the other end
disappears. (In Chapter 29 we’ll see that when NFS uses TCP, both the client and server set this
option. But in Chapter 26 with Rlogin and Telnet, only the servers set the option, not the clients.)

If there is no activity on a given connection for 2 hours, the server sends a probe segment to the client.
(We’ll see what the probe segment looks like in the examples that follow.) The client host must be in
one of four states.

1. The client host is still up and running and reachable from the server. The client’s TCP responds
normally and the server knows that the other end is still up. The server’s TCP will reset the
keepalive timer for 2 hours in the future. If there is application traffic across the connection
before the next 2-hour timer expires, the timer is reset for 2 hours in the future, following the
exchange of data.
2. The client’s host has crashed and is either down or in the process of rebooting. In either case, its
TCP is not responding. The server will not receive a response to its probe and it times out after 75
seconds. The server sends a total of 10 of these probes, 75 seconds apart, and if it doesn’t receive
a response, the server considers the client’s host as down and terminates the connection.
3. The client’s host has crashed and rebooted. Here the server will receive a response to its
keepalive probe, but the response will be a reset, causing the server to terminate the connection.
4. The client’s host is up and running, but unreachable from the server. This is the same as scenario
2, because TCP can’t distinguish between the two. All it can tell is that no replies are received to
its probes.

The server does not have to worry about the client’s host being shut down and then rebooted. (This
refers to an operator shutdown, instead of the host crashing.) When the system is shut down by an
operator, all application processes are terminated (i.e., the client process), which causes the client’s
TCP to send a FIN on the connection. Receiving the FIN would cause the server’s TCP to report an
end-of-file to the server process, allowing the server to detect this scenario.

In the first scenario the server application has no idea that the keepalive probes are taking place.
Everything is handled at the TCP layer. It’s transparent to the application until one of scenarios 2, 3, or
4 occurs. In these three scenarios an error is returned to the server application by its TCP. (Normally
the server has issued a read from the network, waiting for data from the client. If the keepalive feature
returns an error, it is returned to the server as the return value from the read.) In scenario 2 the error is
something like "connection timed out," and in scenario 3 we expect "connection reset by peer." The
fourth scenario may look like the connection timed out, or may cause another error to be returned,
depending on whether an ICMP error related to the connection is received. We look at all four
scenarios in the next section.

A perpetual question by people discovering the keepalive option is whether the 2-hour idle time value can be changed.
They normally want it much lower, on the order of minutes. As we show in Appendix E, the value can usually be changed,
but in all the systems described in this appendix, the keepalive interval is a system-wide value, so changing it affects all
users of the option.

The Host Requirements RFC says that an implementation may provide the keepalive feature, but it must not be enabled
unless the application specifically says so. Also, the keepalive interval must be configurable, but it must default to no less
than 2 hours.

320

23.3 Keepalive Examples
We’ll now go through scenarios 2, 3, and 4 from the previous section, to see the packets exchanged
using the keepalive option.

Other End Crashes
Let’s see what happens when the server host crashes and does not reboot. To simulate this we’ll do the
following steps:

Establish a connection between a client (our sock program on the host bsdi) and the standard
echo server on the host svr4. The client enables the keepalive option with the -K option.
Verify that data can go across the connection.
Watch the client’s TCP send keepalive packets every 2 hours, and see them acknowledged by the
server’s TCP.
Disconnect the Ethernet cable from the server, and leave it off until the example is complete. This
makes the client think the server host has crashed.
We expect the client to send 10 keepalive probes, 75 seconds apart, before declaring the
connection dead.

Here is the interactive output on the client:

bsdi % sock -K svr4 echo -K for keepalive option
hello, world type this at beginning, to verify connection is up
hello, world and see this echoed
disconnect Ethernet cable after 4 hours
read error: Connection timed this happens about 6 hours and 10 minutes after
out start

Figure 23.1 shows the tcpdump output. (We have removed the connection establishment and the
window advertisements.)

321

bsdi.1055 > bsdi.echo : P 1:14(13)
1 0.0
ack 1
bsdi.echo > bsdi.1055 : P 1:14(13)
2 0.006105 ( 0.0061)
ack 14
3 0.093140 ( 0.0870) bsdi.1055 > svr4 .echo : . ack 14
7199.972793
4 arp who-has svr4 tell bsdi
(7199.8797)
5 7199.974878 ( 0.0021) arp reply svr4 is-at 0:0:c0: c2:9b:26
6 7199.975741 ( 0.0009) bsdi.1055 > bsdi.echo : . ack 14
7 7199.979843 ( 0.0041) svr4 .echo > bsdi.1055 : . ack 14
14400.134330
(7200.1545)
9 14400.136452 ( 0.0021) arp reply svr4 is-at 0:0:c0: c2:9b:26
10 14400.137391 ( 0.0009) bsdi.1055 > bsdi.echo : . ack 14
11 14400.141408 ( 0.0040) bsdi.echo > bsdi.1055 : . ack 14
21600.318309
(7200.1769)
21675.320373 (
75.0021)
21750.322407 (
75.0020)
21825.324460 (
75.0021)
21900.436749 (
75.1123)
21975.438787 (
75.0020)
22050.440842 (
75.0021)
22125.432883 (
74.9920)
22200.434697 (
75.0018)
22275.436788 (
75.0021)

322

Figure 23.1 Keepalive packets that determine that a host has crashed.

Lines 1, 2, and 3 send the line "hello, world" from the client to the server and back. The first keepalive
probe occurs 2 hours (7200 seconds) later on line 4. The first thing we see is an ARP request and an
ARP reply, before the TCP segment on line 6 can be sent. The keepalive probe on line 6 elicits a
response from the other end (line 7). The same sequence of packets is exchanged 2 hours later in lines
8-11.

If we could see all the fields in the keepalive probes, lines 6 and 10, we would see that the sequence
number field is one less than the next sequence number to be sent (i.e., 13 in this example, when it
should be 14), but because there is no data in the segment, tcpdump does not print the sequence
number field. (It only prints the sequence number for empty segments that contain the SYN, FIN, or
RST flags.) It is the receipt of this incorrect sequence number that forces the server’s TCP to respond
with an ACK to the keepalive probe. The response tells the client the next sequence number that the
server is expecting (14).

Some older implementations based on 4.2BSD do not respond to these keepalive probes unless the segment contains data.
Some systems can be configured to send one garbage byte of data in the probe to elicit the response. The garbage byte
causes no harm, because it’s not the expected byte (it’s a byte that the receiver has previously received and acknowledged),
so it’s thrown away by the receiver. Other systems send the 4.3BSD-style segment (no data) for the first half of the probe
period, and if no response is received, switch to the 4.2BSD-style segment for the last half.

We then disconnect the cable and expect the next probe, 2 hours later, to fail. When this next probe
takes place, notice that we never see the TCP segments on the cable, because the host is not
responding to ARP requests. We can still see that the client sends 10 probes, spaced 75 seconds apart,
before giving up. We can see from our interactive script that the error code returned to the client
process by TCP gets translated into "Connection timed out," which is what happened.

Other End Crashes and Reboots
In this example we’ll see what happens when the client crashes and reboots. The initial scenario is the
same as before, but after we verify that the connection is up, we disconnect the server from the
Ethernet, reboot it, and then reconnect it to the Ethernet. We expect the next keepalive probe to
generate a reset from the server, because the server now knows nothing about this connection. Here is
the interactive session:

bsdi % sock -K svr4 echo -K to enable keepalive option
hi there type this to verify connection is up
hi there and this is echoed back from other end
here server is rebooted while disconnected from
Ethernet
read error: Connection reset by
peer

Figure 23.2 shows the tcpdump output. (We have removed the connection establishment and the
window advertisements.)

323

1 0.0 bsdi.1057 > svr4.echo: P 1:10(9) ack 1
2 0.006406 ( 0.0064) svr4.echo > bsdi.1057: P 1:10(9) ack 10
3 0.176922 ( 0.1705) bsdi.1057 > svr4.echo: . ack 10
7200.067151
(7199.8902)
7200.069751 (
5 arp reply svr4 is-at 0:0:c0:c2:9b:26
0.0026)
7200.070468 (
6 bsdi.1057 > svr4.echo: . ack 10
0.0007)
7200.075050 ( svr4.echo > bsdi.1057: R
7
0.0046) 1135563275:1135563275(0)

Figure 23.2 Keepalive example when other host has crashed and rebooted.

We establish the connection and send 9 bytes of data from the client to the server (lines 1-3). Two
hours later the first keepalive probe is sent by the client, and the response is a reset from the server.
"The client application prints the error "Connection reset by peer," which makes sense.

Other End Is Unreachable
In this example the client has not crashed, but is not reachable during the 10-minute period when the
keepalive probes are sent. An intermediate router may have crashed, a phone line may be temporarily
out of order, or something similar.

To simulate this example we’ll establish a TCP connection from our host slip through our dialup SLIP
link to the host vangogh.cs.berkeley.edu, and then take the link down. First, here is the
interactive output:

slip % sock -K vangogh.cs.berkeley.edu echo
testing we type this line
testing and see it echoed
sometime in here the dialup SLIP link is taken down
read error: No route to host

Figure 23.3 shows the tcpdump output that was collected on the router bsdi. (The connection
establishment and window advertisements have been removed.)

324

slip.1056 > vangogh.echo: P 1:9(8)
1 0.0
ack 1
vangogh.echo > slip.1056: P 1:9(8)
2 0.277669 0.424423
ack 9
3 0.2777) 0.1468) slip. 1056 > vangogh. echo: . ack 9
7200.818081
4 slip. 1056 > vangogh. echo: . ack 9
(7200.3937)
5 7201.243046 ( 0.4250) vangogh, echo > slip. 1056: . ack 9
14400.688106
6 slip.1056 > vangogh.echo: . ack 9
(7199.4451)
sun > slip: icmp; net vangogh
7 14400.689261 ( 0.0012)
unreachable
14475.684360 (
8 slip. 1056 > vangogh. echo: . ack 9
74.9951)
sun > slip: icmp: net vangogh
9 14475.685504 ( 0.0011)
unreachable
14 lines deleted
slip.1056 > vangogh.echo: R 9:9(0)
24 15075.759603 (75.1008)
ack 9
sun > slip: icmp: net vangogh
25 15075.760761 ( 0.0012)
unreachable

Figure 23.3 Keepalive example when other end is unreachable.

We start the example the same as before: lines 1-3 verify that the connection is up. The first keepalive
probe 2 hours later is fine (lines 4 and 5), but before the next one occurs in another 2 hours, we bring
down the SLIP connection between the routers sun and netb. (Refer to the inside front cover for the
topology.)

The keepalive probe in line 6 elicits an ICMP network unreachable from the router sun. As we
described in Section 21.10, this is just a soft error to the receiving TCP on the host slip. It records that
the ICMP error was received, but the receipt of the error does not take down the connection. Nine
more keepalive probes are sent, 75 seconds apart, before the sending host gives up. The error returned
to the application generates a different message this time: "No route to host." We saw in Figure 6.12
that this corresponds to the ICMP network unreachable error.

23.4 Summary
As we said earlier, the keepalive feature is controversial. Protocol experts continue to debate whether
it belongs in the transport layer, or should be handled entirely by the application.

325

It operates by sending a probe packet across a connection after the connection has been idle for 2
hours. Four different scenarios can occur: the other end is still there, the other end has crashed, the
other end has crashed and rebooted, or the other end is currently unreachable. We saw each of these
scenarios with an example, and saw different errors returned for the last three conditions.

In the first two examples that we looked at, had this feature not been provided, and without any
application-level timer, our client would never have known that the other end had crashed, or crashed
and rebooted. In the final example, however, nothing was wrong with the other end, the connection
between them was temporarily down. We must be aware of this limitation when using keepalives.

Exercises
23.1 List some advantages of the keepalive feature.

23.2 List some disadvantages of the keepalive feature.

326

TCP Futures and Performance
24.1 Introduction
TCP has operated for many years over data links ranging from 1200 bits/sec dialup SLIP links to
Ethernets. Ethernets were the predominant form of data link for TCP/IP in the 1980s and early 1990s.
Although TCP operates correctly at speeds higher than an Ethernet (T3 phone lines, FDDI, and gigabit
networks, for example), certain TCP limits start to be encountered at these higher speeds.

This chapter looks at some proposed modifications to TCP that allow it to obtain the maximum
throughput at these higher speeds. We first look at the path MTU discovery mechanism, which we’ve
seen earlier in the text, focusing this time on how it operates with TCP. This often lets TCP use an
MTU greater than 536 for nonlocal connections, increasing its throughput.

We then look at long fat pipes, networks that have a large bandwidth-delay product, and the TCP
limits that are encountered on these networks. Two new TCP options are described that deal with long
fat pipes: a window scale option (to increase TCP’s maximum window above 65535 bytes) and a
timestamp option. This latter option lets TCP perform more accurate RTT measurement for data
segments, and also provides protection against wrapped sequence numbers, which can occur at high
speeds. These two options are defined in RFC 1323 [Jacobson, Braden, and Borman 1992].

We also look at the proposed T/TCP, modifications to TCP for transactions. The transaction mode of
communication features a client request responded to by a server reply. It is a common paradigm for
client-server computing. The goal of T/TCP is to reduce the number of segments exchanged by the
two ends, avoiding the three-way handshake and the four segments to close the connection, so that the
client receives the server’s reply in one RTT plus the time required to process the request.

What is impressive about these new options-path MTU discovery, the window scale option, the
tirnestamp option, and T/TCP - is that they are backward compatible with existing TCP
implementations. Newer systems that include these options can still interoperate with all older
systems. With the exception of an additional field in an ICMP message that can be used by path MTU
discovery, these newer options need only be implemented on the end systems that want to take
advantage of them.

We finish the chapter by looking at recently published figures dealing with TCP performance.

24.2 Path MTU Discovery
In Section 2.9 we described the concept of the path MTU. It is the minimum MTU on any network that
is currently in the path between two hosts. Path MTU discovery entails setting the "don’t fragment"
(DF) bit in the IP header to discover if any router on the current path needs to fragment IP datagrams
that we send. In Section 11.6 we showed the ICMP unreachable error returned by a router that is asked
to forward an IP datagram with the DF bit set when the MTU is less than the datagram size. In Section
11.7 we showed a version of the traceroute program that used this mechanism to determine the
path MTU to a destination. In Section 11.8 we saw how UDP handled path MTU discovery. In this
section we’ll examine how this mechanism is used by TCP, as specified by RFC 1191 [Mogul and
Deering 1990].

327

Of the various systems used in this text (see the Preface) only Solaris 2.x supports path MTU discovery.

TCP’s path MTU discovery operates as follows. When a connection is established, TCP uses the
minimum of the MTU of the outgoing interface, or the MSS announced by the other end, as the
starting segment size. Path MTU discovery does not allow TCP to exceed the MSS announced by the
other end. If the other end does not specify an MSS, it defaults to 536. It is also possible for an
implementation to save path MTU information on a per-route basis, as we mentioned in Section 21.9.

Once the initial segment size is chosen, all IP datagrams sent by TCP on that connection have the DF
bit set. If an intermediate router needs to fragment a datagram that has the DF bit set, it discards the
datagram and generates the ICMP "can’t fragment" error we described in Section 11.6.

If this ICMP error is received, TCP decreases the segment size and retransmits. If the router generated
the newer form of this ICMP error, the segment size can be set to the next-hop MTU minus the sizes
of the IP and TCP headers. If the older ICMP error is returned, the probable value of the next smallest
MTU (Figure 2.5) must be tried. When a retransmission caused by this ICMP error occurs, the
congestion window should not change, but slow start should be initiated.

Since routes can change dynamically, when some time has passed since the last decrease of the path
MTU, a larger value (up to the minimum of the MSS announced by the other end, or the outgoing
interface MTU) can be tried. RFC 1191 recommends this time interval be about 10 minutes. (We saw
in Section 11.8 that Solaris 2.2 uses a 30-sec-ond timer for this.)

Given the normal default MSS of 536 for nonlocal destinations, path MTU discovery avoids
fragmentation across intermediate links with an MTU of less than 576 (which is rare). It can also avoid
fragmentation on local destinations when an intermediate link (e.g., an Ethernet) has a smaller MTU
than the end-point networks (e.g., a token ring). But for path MTU discovery to be more useful, and
take advantage of wide area networks with MTUs greater than 576, implementations must stop using a
default MSS of 536 bytes for nonlocal destinations. A better choice for the MSS is the MTU of the
outgoing interface (minus the size of the IP and TCP headers, of course). (In Appendix E we’ll see that
most implementations allow the system administrator to change this default MSS value.)

An Example
We can see how path MTU discovery operates when an intermediate router has an MTU less than
either of the end point’s interface MTUs. Figure 24.1 shows the topology for this example.

[IMAGE]
Figure 24.1 Topology for path MTU example.

We’ll establish a connection from the host solaris (which supports the path MTU discovery
mechanism) to the host slip. This setup is identical to the one used for our UDP path MTU
discovery example (Figure 11.13) but here we have set the MTU of the interface on slip to 552,
instead of its normal 296. This causes slip to announce an MSS of 512. But leaving the MTU of the
SLIP link on bsdi at 296 will cause TCP segments greater than 256 to be fragmented, and we can see
how the path MTU discovery mechanism on solaris handles this.

We’ll run our sock program on solaris and perform one 512-byte write to the discard server on
slip:

328

solaris % sock -i -nl -w512 slip discard

Figure 24.2 shows the tcpdump output, collected on the SLIP interface on the host sun.

solaris.33016 > slip.discard: S
1 0.0 1171660288:1171660288(0)
win 8760 <mss 1460> (DF)
slip.discard > solaris.33016: S
0.101597
2 137984001:137984001(0)
(0.1016)
ack 1171660289 win 4096 <mss 512>
0.630609 solaris.33016 > slip.discard: P 1:513(512)
3
(0.5290) ack 1 win 9216 (DF)
0.634433
(0.0038)
(DF)
0.660331 solaris.33016 > slip.discard: F 513:513(0)
5
(0.0259) ack 1 win 9216 (DF)
0.752664 slip.discard > solaris.33016: . ack 1 win
6
(0.0923) 4096
1.110342 solaris.33016 > slip.discard: P 1:257(256)
7
(0.3577) ack 1 win 9216 (DF)
8
(0.3290) 3840
1.770154 solaris.33016 > slip.discard: FP 257:513(256)
9
(0.3308) ack 1 win 9216 (DF)
10
(0.3258) 3840
2.138193 slip.discard > solaris.33016: F 1:1(0) ack
11
(0.0422) 514 win 4096
2.310103 solaris.33016 > slip.discard: . ack 2 win
12
(0.1719) 9216 (DF)

Figure 24.2 tcpdump output for path MTU discovery.

The MSS values in lines 1 and 2 are what we expect. We then see solaris send a 512-byte segment
(line 3) containing the 512 bytes of data and the ACK of the SYN. (We saw this combination of the
ACK of a SYN along with the first segment of data in Exercise 18.9.) This generates the ICMP error
in line 4 and we see that the router bsdi generates the newer ICMP error containing the MTU of the
outgoing interface.

It appears that before this error makes it back to solaris, the FIN is sent (line 5). Since slip never
received the 512 bytes of data discarded by the router bsdi, it is not expecting this sequence number
(513), so it responds in line 6 with the expected sequence number (1).

329

At this time the ICMP error has made it back to solaris and it retransmits the 512 bytes of data in
two 256-byte segments (lines 7 and 9). Both are sent with the DF bit set, since there could be another
router beyond bsdi with a smaller MTU.

A longer transfer was run (taking about 15 minutes) and after moving from the 512-byte initial
segment to 256-byte segments, solaris never tried the higher segment size again.

Big Packets or Small Packets?
Conventional wisdom says that bigger packets are better [Mogul 1993, Sec. 15.2.8] because sending
fewer big packets "costs less" than sending more smaller packets. (This assumes the packets are not
large enough to cause fragmentation, since that introduces another set of problems.) The reduced cost
is that associated with the network (packet header overhead), routers (routing decisions), and hosts
(protocol processing and device interrupts). Not everyone agrees with this [Bellovin 1993].

Consider the following example. We send 8192 bytes through four routers, each connected with a Tl
telephone line (1,544,000 bits/sec). First we use two 4096-byte packets, as shown in Figure 24.3.

[IMAGE]
Figure 24.3 Sending two 4096-byte packets through four routers.

The basic problem is that routers are store-and-forward devices. They normally receive the entire input
packet, validate the IP header including the IP checksum, make their routing decision, and start
sending the output packet. In this figure we’re assuming the ideal case where it takes no time for these
operations to occur at the router (the horizontal dashed lines). Nevertheless, it takes four units of time
to send all 8192 bytes from Rl to R4. The time for each hop is

(4096 + 40 bytes) x 8 bits/bytes / 1’544’000 bits/sec = 21.4 ms per hop

(4096 + 40 bytes) x 8 bits/bytes
= 21.4 ms per hop
1,544,000 bits/sec

(We account for the 40 bytes of IP and TCP header.) The total time to send the data is the number of
packets plus the number of hops, minus one, which we can see visually in this example is four units of
time, or 85.6 ms. Each link is idle for two units of time, or 42.8 ms. Figure 24.4 shows what happens if
we send sixteen 512-byte packets.

[IMAGE]
Figure 24.4 Sending sixteen 512-byte packets through four routers.

It takes more units of time, but the units are shorter since a smaller packet is being sent.

(512 + 40 bytes) x 8 bits/byte / 1,544,000 bits/sec = 2.9 ms per hop

The total time is now (18 x 2.9) = 52.2 ms. Each link is again idle for two units of time, which is now
5.8 ms.

330

In this example we have ignored the time required for the ACKs to be returned, the connection
establishment and termination times, and the possible sharing of the links with other traffic.
Nevertheless, measurements in [Bellovin 1993] indicate that bigger is not always better. More research
is required in this area on various networks.

24.3 Long Fat Pipes
In Section 20.7 we showed the capacity of a connection as

capacity (bits) = bandwidth (bits/sec) x round-trip time (sec)

and called this the bandwidth-delay product. This is also called the size of the pipe between the end
points.

Existing limits in TCP are being encountered as this product increases to larger and larger values.
Figure 24.5 shows some values for various types of networks.

Bandwidth Round-trip Bandwidth-delay
Network
(bits/sec) time (ms) product (bytes)
Ethernet LAN 10,000,000 3 3,750
T1 telephone line, transcontinental 1,544,000 60 11,580
T1 telephone line, satellite 1,544,000 500 95,500
T3 telephone line, transcontinental 45,000,000 60 337,500
gigabit, transcontinental 1,000,000,000 60 7,500,000

Figure 24.5 Bandwidth-delay product for various networks.

We show the bandwidth-delay product in bytes, because that’s how we typically measure the buffer
sizes and window sizes required on each end.

Networks with large bandwidth-delay products are called long fat networks (LFNs, pronounced
"elefan(t)s"), and a TCP connection operating on an LFN is called a long fat pipe. Going back to
Figure 20.11 and Figure 20.12, the pipe can be stretched in the horizontal direction (a longer RTT), or
stretched in the vertical direction (a higher bandwidth), or both. Numerous problems are encountered
with long fat pipes.

1. The TCP window size is a 16-bit field in the TCP header, limiting the window to 65535 bytes. As
we can see from the final column in Figure 24.5, existing networks already require a larger
window than this, for maximum throughput.

The window scale option described in Section 24.4 solves this problem.

2. Packet loss in an LFN can reduce throughput drastically. If only a single segment is lost, the fast
retransmit and fast recovery algorithm that we described in

Section 21.7 is required to keep the pipe from draining. But even with this algorithm, the loss of
more than one packet within a window typically causes the pipeline to drain. (If the pipe drains,
slow start gets things going again, but that takes multiple round-trip times to get the pipe filled
again.)

331

Selective acknowledgments (SACKs) were proposed in RFC 1072 [Jacobson and Braden 1988]
to handle multiple dropped packets within a window. But this feature was omitted from RFC 1323,
because the authors felt several technical problems needed to be worked out before including them in
TCP.

3. We saw in Section 21.4 that many TCP implementations only measure one round-trip time per
window. They do not measure the RTT of every segment. Better RTT measurements are required for
operating on an LFN.

The timestamp option, which we describe in Section 24.5, allows more segments to be timed,
including retransmissions.

4. TCP identifies each byte of data with a 32-bit unsigned sequence number. What’s to prevent a
segment that gets delayed in the network from reappearing at a later time, after the connection that it
was associated with has been terminated, and after a new connection has been established between the
same two hosts and port numbers? First recall that the TTL field in the IP header puts an upper bound
on the lifetime of any IP datagram-255 hops or 255 seconds, whichever comes first. In Section
18.6 we defined the maximum segment lifetime (MSL) as an implementation parameter used to
prevent this scenario from happening. The recommended value of the MSL is 2 minutes (giving a
2MSL of 240 seconds), but we saw in Section 18.6 that many implementations use an MSL value
of 30 seconds.

A different problem with TCP’s sequence numbers appears with LFNs. Since the sequence
number space is finite, the same sequence number is reused after 4,294,967,296 bytes have been
transmitted. What if a segment containing the byte with a sequence number N gets delayed in the
network and then reappears later, while the connection is still up? This is only a problem if the same
sequence number N is reused within the MSL period, that is, if the network is so fast that
sequence number wrap occurs in less than MSL. On an Ethernet it takes almost 60 minutes to send this
much data, so there is no chance of this happening, but the time required for the wrap to occur drops as
the bandwidth increases: a T3 telephone line (45 Mbits/sec) wraps in 12 minutes, FDDI (100
Mbits/sec) in 5 minutes, and a gigabit network (1000 Mbits/sec) in 34 seconds. The problem here is
not the bandwidth-delay product, but the bandwidth itself.

In Section 24.6 we describe a way to handle this: the PAWS algorithm (protection against
wrapped sequence numbers), which uses the TCP timestamp option.

4.4BSD contains all the options and algorithms that we describe in the following sections: the window scale option,
the timestamp option, and the protection against wrapped sequence numbers. Numerous vendors are also starting to support
these options.

Gigabit Networks
When networks reach gigabit speeds, things change. [Partridge 1994] covers gigabit networks in
detail. Here we’ll look at the differences between latency and bandwidth [Kleinrock 1992].

Consider sending a one million byte file across the United States, assuming a 30-ms latency. Figure
24.6 shows two scenarios, the top illustration uses a Tl telephone line (1,544,000 bits/sec) and the
bottom uses a 1 gigabit/sec network. Time is shown along the x-axis, with the sender on the left and
the receiver on the right, and capacity on the y-axis. The shaded area in both pictures is the one million
bytes to send.

332

[IMAGE]
Figure 24.6 Sending a 1-Mbyte file across networks with a 30-ms latency.

Figure 24.6 shows the status of both networks after 30 ms. With both networks the first bit of data
reaches the other end after 30 ms (the latency), but with the T1 network the capacity of the pipe is only
5,790 bytes, so 994,210 bytes are still at the sender, waiting to be sent. The capacity of the gigabit
network, however, is 3,750,000 bytes, so the entire file uses just over 25% of the pipe. The last bit of
the file reaches the receiver 8 ms after the first bit.

The total time to transfer the file across the T1 network is 5.211 seconds. If we throw more bandwidth
at the problem, a T3 network (45,000,000 bits/sec), the total time decreases to 0.208 seconds.
Increasing the bandwidth by a factor of 29 reduces the total time by a factor of 25.

With the gigabit network the total time to transfer the file is 0.038 seconds: the 30-ms latency plus the
8 ms for the actual file transfer. Assuming we could double the bandwidth to 2 gigabits/sec, we only
reduce the total time to 0.034 seconds: the same 30-ms latency plus 4 ms to transfer the file. Doubling
the bandwidth now decreases the total time by only 10%. At gigabit speeds we are latency limited, not
bandwidth limited.

The latency is caused by the speed of light and can’t be decreased (unless Einstein was wrong). The
effect of this fixed latency becomes worse when we consider the packets required to establish and
terminate a connection. Gigabit networks will cause several networking issues to be looked at
differently.

24.4 Window Scale Option
The window scale option increases the definition of the TCP window from 16 to 32 bits. Instead of
changing the TCP header to accommodate the larger window, the header still holds a 16-bit value, and
an option is defined that applies a scaling operation to the 16-bit value. TCP then maintains the "real"
window size internally as a 32-bit value.

We saw an example of this option in Figure 18.20. The 1-byte shift count is between 0 (no scaling
performed) and 14. This maximum value of 14 is a window of 1,073,725,440 bytes (65535 x 2 14 ).

This option can only appear in a SYN segment; therefore the scale factor is fixed in each direction
when the connection is established. To enable window scaling, both ends must send the option in their
SYN segments. The end doing the active open sends the option in its SYN, but the end doing the
passive open can send the option only if the received SYN specifies the option. The scale factor can be
different in each direction.

If the end doing the active open sends a nonzero scale factor, but doesn’t receive a window scale
option from the other end, it sets its send and receive shift count to 0. This lets newer systems
interoperate with older systems that don’t understand the new option.

The Host Requirements RFC requires TCP to accept an option in any segment. (The only previously defined option, the
maximum segment size, only appeared in SYN segments.) It further requires TCP to ignore any option it doesn’t
understand. This is made easy since all the new options have a length field (Figure 18.20).

Assume we are using the window scale option, with a shift count of S for sending and a shift count of
R for receiving. Then every 16-bit advertised window that we receive from the other end is left shifted
by R bits to obtain the real advertised window size. Every time we send a window advertisement to the
other end, we take our real 32-bit window size and right shift it S bits, placing the resulting 16-bit

333

value in the TCP header.

The shift count is automatically chosen by TCP, based on the size of the receive buffer. The size of
this buffer is set by the system, but the capability is normally provided for the application to change it.
(We discussed this buffer in Section 20.4.)

An Example
If we initiate a connection using our sock program from the 4.4BSD host
vangogh.cs.berkeley.edu, we can see its TCP calculate the window scale factor. The
following interactive output shows two consecutive runs of the program, first specifying a receive
buffer of 128000 bytes, and then a receive buffer of 220000 bytes:

vangogh % sock -v -R128000 bsdi.tuc.noao.edu echo
SO_RCVBUF = 128000
connected on 128.32.130.2.4107 to 140.252.13.35.7
TCP_MAXSEG = 512
hello, world we type this line
hello, world and it’s echoed here
^D type end-of-file character to terminate
vangogh % sock -v -R220000 bsdi.tuc.noao.edu echo
SO_RCVBUF = 220000
connected on 128.32.130.2.4108 to 140.252.13.35.7
TCP_MAXSEG = 512
bye, bye type this line
bye, bye and it’s echoed here
^D type end-of-file character to terminate

Figure 24.7 shows the tcpdump output for these two connections. (We have deleted the final 8 lines
for the second connection, because nothing new is shown.)

334

vangogh.4107 > bsdi.echo: S
462402561:462402561(0)
1 0.0 win 65535
<mss 512,nop,wscale l,nop,nop,timestamp
995351 0>
bsdi.echo > vangogh.4107: S
0.003078 (
2 177032705:177032705(0)
0.0031)
ack 462402562 win 4096 <mss 512>
0.300255 (
3 vangogh.4107 > bsdi.echo: . ack 1 win 65535
0.2972)
16.920087 vangogh.4107 > bsdi.echo: P 1:14(13) ack 1
4
(16.6198) win 65535
16.923063 ( bsdi.echo > vangogh.4107: P 1:14(13) ack 14
5
0.0030) win 4096
17.220114 ( vangogh.4107 > bsdi.echo: . ack 14 win
6
0.2971) 65535
26.640335 ( vangogh.4107 > bsdi.echo: F 14:14(0) ack 14
7
9.4202) win 65535
26.642688 (
8 bsdi.echo > vangogh.4107: . ack 15 win 4096
0.0024)
26.643964 ( bsdi.echo > vangogh.4107: F 14:14(0) ack 15
9
0.0013) win 4096
26.880274 ( vangogh.4107 > bsdi.echo: . ack 15 win
10
0.2363) 65535
vangogh.4108 > bsdi.echo: S
468226561:468226561(0)
44.400239
11 win 65535
(17.5200)
<mss 512,nop,wscale 2,nop,nop,timestamp
995440 0>
bsdi.echo > vangogh.4108: S
44.403358 (
12 182792705:182792705(0)
0.0031)
ack 468226562 win 4096 <mss 512>
44.700027 (
13 vangogh.4108 > bsdi.echo: . ack 1 win 65535
0.2967)
remainder of this connection deleted

Figure 24.7 Example of window scale option.

In line 1 vangogh advertises a window of 65535 and specifies the window scale option with a shift
count of 1. This advertised window is the largest possible value that is less than the receive buffer size
(128000), because the window field in a SYN segment is never scaled.

335

The scale factor of 1 means vangogh would like to send window advertisements up to 131070
(65535 x 2 1 ). This will accommodate our receive buffer size (128000). Since bsdi does not send the
window scale option in its SYN (line 2), the option is not used.

Notice that vangogh continues to use the largest window possible (65535) for the remainder of the
connection.

For the second connection vangogh requests a shift count of 2, meaning it would like to send
window advertisements up to 262140 (65535 x 2 2 ), which is greater than our receive buffer size
(220000).

24.5 Timestamp Option
The timestamp option lets the sender place a timestamp value in every segment. The receiver reflects
this value in the acknowledgment, allowing the sender to calculate an RTT for each received ACK.
(We must say "each received ACK" and not "each segment" since TCP normally acknowledges
multiple segments per ACK.) We said that many current implementations only measure one RTT per
window, which is OK for windows containing eight segments. Larger window sizes, however, require
better RTT calculations.
Section 3.1 of RFC 1323 gives the signal processing reasons for requiring better RTT estimates for bigger windows.
Basically the RTT is measured by sampling a data signal (the data segments) at a lower frequency (once per window). This
introduces aliasing into the estimated RTT. When there are eight segments per window, the sample rate is one-eighth the
data rate, which is tolerable, but with 100 segments per window, the sample rate is 1/IOOth the data rate. This can cause
the estimated RTT to be inaccurate, resulting in unnecessary retransmissions. If a segment is lost, it only gets worse.

Figure 18.20 showed the format of the timestamp option. The sender places a 32-bit value in the first
field, and the receiver echoes this back in the reply field. TCP headers containing this option will
increase from the normal 20 bytes to 32 bytes.

The timestamp is a monotonically increasing value. Since the receiver echoes what it receives, the
receiver doesn’t care what the timestamp units are. This option does not require any form of clock
synchronization between the two hosts. RFC 1323 recommends that the timestamp value increment by
one between 1 ms and 1 second.

4.4BSD increments the timestamp clock once every 500 ms and this timestamp clock is reset to 0 on a reboot.

In Figure 24.7, if we look at the timestamp in segment 1 and the timestamp in segment II, the difference (89 units)
corresponds to 500 ms per unit for the time difference of 44.4 seconds.

The specification of this option during connection establishment is handled the same way as the
window scale option in the previous section. The end doing the active open specifies the option in its
SYN. Only if it receives the option in the SYN from the other end can the option be sent in future
segments.

We’ve seen that a receiving TCP does not have to acknowledge every data segment that it receives.
Many implementations send an ACK for every other data segment. If the receiver sends an ACK that
acknowledges two received data segments, which received timestamp is sent back in the timestamp
echo reply field?

To minimize the amount of state maintained by either end, only a single timestamp value is kept per
connection. The algorithm to choose when to update this value is simple.

336

1. TCP keeps track of the timestamp value to send in the next ACK (a variable named tsrecent) and
the acknowledgment sequence number from the last ACK that was sent (a variable named
lastack). This sequence number is the next sequence number the receiver is expecting.
2. When a segment arrives, if the segment contains the byte numbered lastack, then the timestamp
value from the segment is saved in tsrecent.
3. Whenever a timestamp option is sent, tsrecent is sent as the timestamp echo reply field and the
sequence number field is saved in lastack.

This algorithm handles the following two cases:

1. If ACKs are delayed by the receiver, the timestamp value returned as the echo value will
correspond to the earliest segment being acknowledged.

For example, if two segments containing bytes 1-1024 and 1025-2048 arrive, both with a
timestamp option, and the receiver acknowledges them both with an ACK 2049, the timestamp in
the ACK will be the value from the first segment containing bytes 1-1024. This is correct because
the sender must calculate its retransmission timeout taking the delayed ACKs into consideration.

2. If a received segment is in-window but out-of-sequence, implying that a previous segment has
been lost, when that missing segment is received, its time-stamp will be echoed, not the
timestamp from the out-of-sequence segment.

For example, assume three segments, each containing 1024 bytes, are received in the following
order: segment 1 with bytes 1-1024, segment 3 with bytes 2049-3072, then segment 2 with bytes
1025-2048. The ACKs sent back will be ACK 1025 with the timestamp from segment 1 (a
normal ACK for data that was expected), ACK 1025 with the timestamp from segment 1 (a
duplicate ACK in response to the in-window but the out-of-sequence segment), then ACK 3073
with the timestamp from segment 2 (not the later timestamp from segment 3). This has the effect
of overestimating the RTT when segments are lost, which is better than underestimating it. Also,
if the final ACK contained the timestamp from segment 3, it might include the time required for
the duplicate ACK to be returned and segment 2 to be retransmitted, or it might include the time
for the sender’s retransmission timeout for segment 2 to expire. In either case, echoing the
timestamp from segment 3 could bias the sender’s RTT calculations.

Although the timestamp option allows for better RTT calculations, it also provides a way for the
receiver to avoid receiving old segments and considering them part of the existing data segment. The
next section describes this.

24.6 PAWS: Protection Against Wrapped Sequence Numbers
Consider a TCP connection using the window scale option with the largest possible window, 1
gigabyte (2 30 ). (The largest window is just smaller than this, 65535 x 2 14 , not 2 16 x 2 14 , but that
doesn’t affect this discussion.) Also assume the timestamp option is being used and that the timestamp
value assigned by the sender increments by one for each window that is sent. (This is conservative.
Normally the timestamp increments faster than this.) Figure 24.8 shows the possible data flow
between the two hosts, when transferring 6 gigabytes. To avoid lots of IO-digit numbers, we use the
notation G to mean a multiple of 1,073,741,824. We also use the notation from tcpdump that J:K
means byte 1 through and including byte K - 1.

337

Send Send
Time Bytes sent Receive
sequence# timestamp
A 0G:1G 0G:1G 1 OK
B 1G:2G 1G:2G 2 OK but one segment lost and retransmitted
C 2G:3G 2G:3G 3 OK
D 3G:4G 3G:4G 4 OK
E 4G:5G 0G:1G 5 OK
F 5G:6G 1G:2G 6 OK but retransmitted segment reappears

Figure 24.8 Transferring 6 gigabytes in six 1-gigabyte windows.

The 32-bit sequence number wraps between times D and E. We assume that one segment gets lost at
time B and is retransmitted. We also assume that this lost segment reappears at time F.

This assumes that the time difference between the segment getting lost and reappearing is less than the
MSL; otherwise the segment would have been discarded by some router when its TTL expired. As we
mentioned earlier, it is only with high-speed connections that this problem appears, where old
segments can reappear and contain sequence numbers currently being transmitted.

We can also see from Figure 24.8 that using the timestamp prevents this problem. The receiver
considers the timestamp as a 32-bit extension of the sequence number. Since the lost segment that
reappears at time F has a timestamp of 2, which is less than the most recent valid timestamp (5 or 6), it
is discarded by the PAWS algorithm.

The PAWS algorithm does not require any form of time synchronization between the sender and
receiver. All the receiver needs is for the timestamp values to be mono-tonically increasing, and to
increase by at least one per window.

24.7 T/TCP: A TCP Extension for Transactions
TCP provides a virtual-circuit transport service. There are three distinct phases in the life of a
connection: establishment, data transfer, and termination. Applications such as remote login and file
transfer are well suited to a virtual-circuit service.

Other applications, however, are designed to use a transaction service. A transaction is a client request
followed by a server response with the following characteristics:

1. The overhead of connection establishment and connection termination should be avoided. When
possible, send one request packet and receive one reply packet.
2. The latency should be reduced to RTT plus SPT, where RTT is the round-trip time and SPT is the
server processing time to handle the request.
3. The server should detect duplicate requests and not replay the transaction when a duplicate
request arrives. (Avoiding the replay means the server does not process the request again. The
server sends back the saved reply corresponding to that request.)

338

One application that we’ve already seen that uses this type of service is the Domain Name System
(Chapter 14), although the DNS is not concerned with the server replaying duplicate requests.

Today the choice an application designer has is TCP or UDP. TCP provides too many features for
transactions, and UDP doesn’t provide enough. Usually the application is built using UDP (to avoid
the overhead of TCP connections) but many of the desirable features (dynamic timeout and
retransmission, congestion avoidance, etc.) are placed into the application, where they’re reinvented
over and over again.

A better solution is to provide a transport layer that provides efficient handling of transactions. The
transaction protocol we describe in this section is called T/TCP. Our description is from its definition,
RFC 1379 [Braden 1992b] and [Braden 1992c].

Most TCPs require 7 segments to open and close a connection (see Figure 18.13). Three more
segments are then added: one with the request, another with the reply and an ACK of the request, and
a third with the ACK of the reply. If additional control bits are added onto the segments-that is, the
first segment contains a SYN, the client request, and a FIN-the client still sees a minimal overhead of
twice the RTT plus SPT. (Sending a SYN along with data and a FIN is legal; whether current TCPs
handle it correctly is another question.)

Another problem with TCP is the TIME_WAIT state and its required 2MSL wait. As shown in
Exercise 18.14, this limits the transaction rate between two hosts to about 268 per second.

The two modifications required for TCP to handle transactions are to avoid the three-way handshake
and shorten the TIME_WAIT state. T/TCP avoids the three-way handshake by using an accelerated
open:

1. It assigns a 32-bit connection count (CC) value to connections it opens, either actively or
passively. A host’s CC value is assigned from a global counter that gets incremented by 1 each
time it’s used.

2. Every segment between two hosts using T/TCP includes a new TCP option named CC. This
option has a length of 6 bytes and contains the sender’s 32-bit CC value for the connection.

3. A host maintains a per-host cache of the last CC value received in an acceptable SYN segment
from that host.

4. When a CC option is received on an initial SYN, the receiver compares the value with the cached
value for the sender. If the received CC is greater than the cached CC, the SYN is new and any
data in the segment is passed to the receiving application (the server). The connection is called
half-synchronized.

If the received CC is not greater than the cached CC, or if the receiving host doesn’t have a
cached CC for this client, the normal TCP three-way handshake is performed.

5. The SYN, ACK segment in response to an initial SYN echoes the received CC value in another
new option named CCECHO.

6. The CC value in a non-SYN segment detects and rejects any duplicate segments from previous
incarnations of the same connection.

339

The accelerated open avoids the need for a three-way handshake unless either the client or server has
crashed and rebooted. The cost is that the server must remember the last CC received from each client.

The TIME_WAIT state is shortened by calculating the TIME_WAIT delay dynamically, based on the
measured RTT between the two hosts. The TIME_WAIT delay is set to 8 times RTO, the
retransmission timeout value (Section 21.3).

Using these features the minimal transaction sequence is an exchange of three segments:

1. Client to server, caused by an active open: client-SYN, client-data (the request), client-FIN, and
client-CC.

When the server TCP with the passive open receives this segment, if the client-CC is greater than
the cached CC for this client host, the client-data is passed to the server application, which
processes the request.

2. Server to client: server-SYN, server-data (reply), server-FIN, ACK of client-FIN, server-CC, and
CCECHO of client-CC. Since TCP acknowledgments are cumulative, this ACK of the client FIN
acknowledges the client’s SYN, data, and FIN.

When the client TCP receives this segment it passes the reply to the client application.

3. Client to server: ACK of server-FIN, which acknowledges the server’s SYN, data, and FIN.

The client’s response time to its request is RTT plus SPT.

There are many fine points to the implementation of this TCP option that are covered in the references.
We summarize them here:

The server’s SYN, ACK (the second segment) should be delayed, to allow the reply to piggyback
with it. (Normally the ACK of a SYN is not delayed.) It can’t delay too long, or the client will
time out and retransmit.
The request can require multiple segments, but the server must handle their possible out-of-order
arrival. (Normally when data arrives before the SYN, the data is discarded and a reset is
generated. With T/TCP this out-of-order data should be queued instead.)
The API must allow the server process to send data and close the connection in a single operation
to allow the FIN in the second segment to piggyback with the reply (Normally the application
would write the reply, causing a data segment to be sent, and then close the connection, causing
the FIN to be sent.)
The client is sending data in the first segment before receiving an MSS announcement from the
server. To avoid restricting the client to an MSS of 536, the MSS for a given host should be
cached along with its CC value.
The client is also sending data to the server without receiving a window advertisement from the
server. T/TCP suggests a default window of 4096 bytes and also caching the congestion threshold
for the server.
With the minimal three-segment exchange there is only one RTT that can be measured in each
direction. Plus the client’s measured RTT includes the server’s processing time. This means the
smoothed RTT value and its variance also must be cached for the server, similar to what we
described in Section 21.9.

340

The appealing feature of T/TCP is that it is a minimal set of changes to an existing protocol but allows
backward compatibility with existing implementations. It also takes advantage of existing engineering
features of TCP (dynamic timeout and retransmission, congestion avoidance, etc.) instead of forcing
the application to deal with these issues.

An alternative transaction protocol is VMTP, the Versatile Message Transaction Protocol. It is
described in RFC 1045 [Cheriton 1988]. Unlike T/TCP, which is a small set of extensions to an
existing protocol, VMTP is a complete transport layer that uses IP VMTP handles error detection,
retransmission, and duplicate suppression. It also supports multicast communication.

24.8 TCP Performance
Published numbers in the mid-1980s showed TCP throughput on an Ethernet to be around 100,000 to
200,000 bytes per second. (Section 17.5 of [Stevens 1990] gives these references.) A lot has changed
since then. It is now common for off-the-shelf hardware (workstations and faster personal computers)
to deliver 800,000 bytes or more per second.

It is a worthwhile exercise to calculate the theoretical maximum throughput we could see with TCP on
a 10 Mbits/sec Ethernet [Warnock 1991]. We show the basics for this calculation in Figure 24.9. This
figure shows the total number of bytes exchanged for a full-sized data segment and an ACK.

Data ACK
Field
#bytes #bytes
Ethernet preamble 8 8
Ethernet destination address 6 6
Ethernet source address 6 6
Ethernet type field 2 2
IP header 20 20
TCP header 20 20
user data 1460 0
pad (to Ethernet minimum) 0 6
Ethernet CRC 4 4
interpacket gap (9.6 microsec) 12 12
total 1538 84

Figure 24.9 Field sizes for Ethernet theoretical maximum throughput calculation.

We must account for all the overhead: the preamble, the PAD bytes that are added to the
acknowledgment, the CRC, and the minimum interpacket gap (9.6 microseconds, which equals 12
bytes at 10 Mbits/sec).

We first assume the sender transmits two back-to-back full-sized data segments, and then the receiver
sends an ACK for these two segments. The maximum throughput (user data) is then

throughput = 2 x 1460 bytes / (2 x 1538 + 84 bytes) x 10,000,000 bits/sec / 8 buts/byte =
= 1,155,063 bytes/sec

341

If the TCP window is opened to its maximum size (65535, not using the window scale option), this
allows a window of 44 1460-byte segments. If the receiver sends an ACK every 22nd segment the
calculation becomes

throughput = 22 x 1460 bytes / (22 x 1538 + 84 bytes) x 10,000,000 bits/sec / 8 buts/byte =
= 1,183,667 bytes/sec

This is the theoretical limit, and makes certain assumptions: an ACK sent by the receiver doesn’t
collide on the Ethernet with one of the sender’s segments; the sender can transmit two segments with
the minimum Ethernet spacing; and the receiver can generate the ACK within the minimum Ethernet
spacing. Despite the optimism in these numbers, [Wamock 1991] measured a sustained rate of
1,075,000 bytes/sec on an Ethernet, with a standard multiuser workstation (albeit a fast workstation),
which is within 90% of the theoretical value.

Moving to faster networks, such as FDDI (100 Mbits/sec), [Schryver 1993] indicates that three
commercial vendors have demonstrated TCP over FDDI between 80 and 98 Mbits/sec. When even
greater bandwidth is available, [Borman 1992] reports up to 781 Mbits/sec between two Cray Y-MP
computers over an 800 Mbits/sec HIPPI channel, and 907 Mbits/sec between two processes using the
loopback interface on a Cray Y-MP. The following practical limits apply for any real-world scenario
[Borman 1991].

1. You can’t run any faster than the speed of the slowest link.
2. You can’t go any faster than the memory bandwidth of the slowest machine. This assumes your
implementation makes a single pass over the data. If not (i.e., your implementation makes one
pass over the data to copy it from user space into the kernel, then another pass over the data to
calculate the TCP checksum), you’ll run even slower. [Dalton et al. 1993] describe performance
improvements to the standard Berkeley sources that reduce the number of data copies to one.
[Partridge and Pink 1993] applied the same "copy-and-checksum" change to UDP, along with
other performance improvements, and improved UDP performance by about 30%.
3. You can’t go any faster than the window size offered by the receiver, divided by the round-trip
time. (This is our bandwidth-delay product equation, using the window size as the
bandwidth-delay product, and solving for the bandwidth.) If we use the maximum window scale
factor of 14 from Section 24.4, we have a window size of 1 Gbyte, so this divided by the RTT is
the bandwidth limit.

The bottom line in all these numbers is that the real upper limit on how fast TCP can run is determined
by the size of the TCP window and the speed of light. As concluded by [Partridge and Pink 1993],
many protocol performance problems are implementation deficiencies rather than inherent protocol
limits.

24.9 Summary
This chapter has looked at five new TCP features: path MTU discovery, window scale option,
timestamp option, protection against wrapped sequence numbers, and improved transactional
processing using TCP. We saw that the middle three features are required for optimal performance on
long fat pipes-networks with a large bandwidth-delay product.

Path MTU discovery allows TCP to use windows larger than the default of 536 for nonlocal
connections, when the path MTU is larger. This can improve performance.

342

The window scale option takes the maximum TCP window size from 65535 bytes to just over 1
gigabyte. The timestamp option lets more segments be accurately timed, and also lets the receiver
provide protection against wrapped sequence numbers (PAWS). This is essential for high-speed
connections. These new TCP options are negotiated at connection establishment, and ignored by older
systems that don’t understand them, allowing newer systems to interoperate with older systems.

The TCP extensions for transactions, T/TCP, allow a client-server request-reply sequence to be
completed using only three segments in the usual case. It avoids the three-way handshake and shortens
the TIME_WAIT state by caching a small amount of information for each host with which it has
established a connection. It also overloads the data segments with the SYN and FIN flags.

We finished the chapter with a look at TCP performance, since there is still much inaccurate folklore
about how fast TCP can run. For a well-tuned implementation using the newer features described in
this chapter, TCP performance is limited only by the maximum 1-gigabyte window and the speed of
light (i.e., the round-trip time).

Exercises
24.1 What does it mean when a system sends an initial SYN segment with a window scale factor of 0?

24.2 If the host bsdi in Figure 24.7 supported the window scale option, what is the expected value of
the 16-bit window size field in the TCP header from vangogh in segment 3? Similarly, if the option
were in use for the second connection in that figure, what would be the advertised window in segment
13?

24.3 Instead of fixing the window scale factor when the connection is established, could the window
scale option have been defined to also appear when the scaling factor changes?

24.4 At what data rate does sequence number wrap become a problem, assuming an MSL of 2
minutes?

24.5 PAWS is defined to operate within a single connection only. What modifications would have to
be made to TCP to use PAWS as a replacement for the 2MSL wait (the TIME_WAIT state)?

24.6 In our example at the end of Section 24.4, why did our sock program output the size of the
receive buffer before the line that followed (with the IP addresses and port numbers)?

24.7 Redo the calculations of the throughput in Section 24.8 assuming an MSS of 1024.

24.8 How does the timestamp option affect Karn’s algorithm (Section 21.3)?

24.9 If TCP sends data with the SYN segment that’s generated by an active open (without using the
extensions we described in Section 24.7), what does the receiving TCP do with the data?

24.10 In Section 24.7 we said that without the T/TCP extensions, even if the active open is sent with
data and a FIN, the client delay in receiving the server’s response is still twice the RTT plus SPT.
Show the segments to account for this.

24.11 Redo Exercise 18.14 assuming T/TCP support and the minimum RTO supported by
Berkeley-derived systems of one-half second.

343

24.12 If we implement T/TCP and measure the transaction time between two hosts, what can we
compare it to, to determine its efficiency?

344

SNMP: Simple Network Management Protocol
25.1 Introduction
As the number of networks within an organization grows, along with the diversity of systems
comprising this internet (routers from various vendors, hosts with embedded router functionality,
terminal servers, etc.), managing all these systems within a coherent framework becomes important.
This chapter looks at the standards used within the Internet protocol suite for network management.

Network management of a TCP/IP internet consists of network management stations (managers)
communicating with network elements. The network elements can be anything that runs the TCP/IP
protocol suite: hosts, routers, X terminals, terminal servers, printers, and so on. The software in the
network element that runs the management software is called the agent. Management stations are
normally workstations with color monitors that graphically display relevant facts about the elements
being monitored (which links are up and down, volume of traffic across various links over time, etc.).

"The communication can be two way: the manager asking the agent for a specific value ("how many
ICMP port unreachables have you generated?"), or the agent telling the manager that something
important happened ("an attached interface has gone down"). Also, the manager should be able to set
variables in the agent ("change the value of the default IP TTL to 64"), in addition to reading variables
from the agent. TCP/IP network management consists of three pieces.

1. A Management Information Base (MIB) that specifies what variables the network elements
maintain (the information that can be queried and set by the manager). RFC 1213 [McCloghrie
and Rose 1991] defines the second version of this, called MIB-II.
2. A set of common structures and an identification scheme used to reference the variables in the
MIB. This is called the Structure of Management Information (SMI) and is specified in RFC
1155 [Rose and McCloghrie 1990]. For example, the SMI specifies that a Counter is a
nonnegative integer that counts from 0 through 4,294,967,295 and then wraps around to 0.
3. The protocol between the manager and the element, called the Simple Network Management
Protocol (SNMP). RFC 1157 [Case et al. 1990] specifies the protocol. This details the format of
the packets exchanged. Although a wide variety of transport protocols could be used, UDP is
normally used with SNMP.

These RFCs define what is now called SNMPv1, or just SNMP, which is the topic of this chapter.
During 1993 additional RFCs were published specifying SNMP Version 2 (SNMPv2), which we
describe in Section 25.12.

Our approach to SNMP in this chapter is to describe the protocol between the manager and the agent
first, and then look at the data types for the variables maintained by the agent. We describe the
database of information maintained by the agent (the MIB), looking at the groups that we’ve described
in this text: IP, UDP, TCP, and so on. We show examples at each point along the way, tying network
management back to the protocol concepts from earlier chapters.

25.2 Protocol
SNMP defines only five types of messages that are exchanged between the manager and agent.

345

1. Fetch the value of one or more variables: the get-request operator.
2. Fetch the next variable after one or more specified variables: the get-next-request
operator. (We describe what we mean by "next" later in this chapter.)
3. Set the value of one or more variables: the set-request operator.
4. Return the value of one or more variables: the get-response operator. This is the message
returned by the agent to the manager in response to the get-request, get-next-request,
and set-request operators.
5. Notify the manager when something happens on the agent: the trap operator.

The first three messages are sent from the manager to the agent, and the last two are from the agent to
the manager. (We’ll refer to the first three as the get, get-next, and set operators.) Figure 25.1
summarizes these five operators.

Since four of the five SNMP messages are simple request-reply protocols (the manager sends a
request, the agent sends back a reply) SNMP uses UDP. This means that a request from the manager
may not arrive at the agent, and the agent’s reply may not make it back to the manager. The manager
probably wants to implement a timeout and retransmission.

[IMAGE]
Figure 25.1 Summary of the five SNMP operators.

The manager sends its three requests to UDP port 161. The agent sends traps to UDP port 162. By
using two different port numbers, a single system can easily run both a manager and an agent. (See
Exercise 25.1.)

Figure 25.2 shows the format of the five SNMP messages, encapsulated in a UDP datagram.

[IMAGE]
Figure 25.2 Format of the five SNMP messages.

In this figure we specify the size in bytes of the IP and UDP headers only. This is because the
encoding used for the SNMP message-called ASN.1 and BER, which we describe later in this
chapter-varies depending on the type of variable and its value.

The version is 0. This value is really the version number minus one, as the version of SNMP that we
describe is called SNMPv1.

Figure 25.3 shows the values for the PDU type. (PDU stands for Protocol Data Unit, a fancy word for
"packet.")

PDU type Name
0 get-request
1 get-next-request
2 set-request
3 get-response
4 trap

346

Figure 25.3 PDU types for SNMP messages.

The community is a character string that is a cleartext password between the manager and agent. A
common value is the 6-character string public.

For the get, get-next, and set operators, the request ID is set by the manager, and returned by
the agent in the get-response message. We’ve seen this type of variable with other UDP
applications. (Recall the DNS identification field in Figure 14.3, and the transaction ID field in Figure
16.2.) It lets the client (the manager in this case) match the responses from the server (the agent) to the
queries that the client issued. "This field also allows the manager to issue multiple requests to one or
more agents, and then be able to sort out the returned replies.

The error status is an integer returned by the agent specifying an error. Figure 25.4 shows the values,
names, and descriptions.

error status Name Description
0 noError all is OK
1 tooBig agent could not fit reply into a single SNMP message
2 noSuchName operation specified a nonexistent variable
3 badValue a set operation specified an invalid value or syntax
4 readonly manager tried to modify a read-only variable
5 genErr some other error

Figure 25.4 SNMP error status values.

If an error occurred, the error index is an integer offset specifying which variable was in error. It is set
by the agent only for the noSuchName, badValue, and readonly errors.

A list of variable names and values follows in the get, get-next, and set requests. The value
portion is ignored for the get and get-next operators.

For the trap operator (a PDU type of 4), the format of the SNMP message changes. We describe the
fields in the trap header when we describe this operator in Section 25.10.

25.3 Structure of Management Information
SNMP uses only a few different types of data. In this section we’ll look at those data types, without
worrying about how the data is actually encoded (that is, the bit pattern used to store the data).

INTEGER. Some variables are declared as an integer with no restrictions (e.g., the MTU of an
interface), some are defined as taking on specific values (e.g., the IP forwarding flag is 1 if
forwarding is enabled or 2 if forwarding is disabled), and others are defined with a minimum and
maximum value (e.g., UDP and TCP port numbers are between 0 and 65535).
OCTET STRING. A string of 0 or more 8-bit bytes. Each byte has a value between 0 and 255. In
the BER encoding used for this data type and the next, a count of the number of bytes in the
string precedes the string. These strings are not null-terminated strings.
DisplayString. A string of 0 or more 8-bit bytes, but each byte must be a character from the

347

NVT ASCII set (Section 26.4). All variables of this type in the MIB-II must contain no more than
255 characters. (A O-length string is OK.)
OBJECT IDENTIFIER. We describe these in the next section.
NULL. This indicates that the corresponding variable has no value. It is used, for example, as the
value of all the variables in a get or get-next request, since the values are being queried, not set.
lpAddress. An OCTET STRING of length 4, with 1 byte for each byte of the IP address.
PhysAddress. An OCTET STRING specifying a physical address (e.g., a 6-byte Ethernet
address).
Counter. A nonnegative integer whose value increases monotonically from 0 to 2 32 -1
(4,294,967,295), and then wraps back to 0.
Gauge. A nonnegative integer between 0 and 2 32 -1, whose value can increase or decrease, but
latches at its maximum value. That is, if the value increments to 2 32 - 1, it stays there until reset.
The MIB variable tcpCurrEstab is an example: it is the number of TCP connections currently
in the ESTABLISHED or CLOSE_WAIT state.
TimeTicks. A counter that counts the time in hundredths of a second since some epoch.
Different variables can specify this counter from a different epoch, so the epoch used for each variable
of this type is specified when the variable is declared in the MIB. For example, the variable
sysUpTime is the number of hundredths of a second that the agent has been up.
SEQUENCE. This is similar to a structure in the C programming language. For example, we’ll see
that the MIB defines a SEQUENCE named UdpEntry containing information about an agent’s
active UDP end points. (By "active" we mean ports currently in use by an application.) Two entries are
in the structure:

1. udpLocalAddress, of type lpAddress, containing the local IP address.
2. udpLocalPort, of type INTEGER, in the range 0 through 65535, specifying the local port
number.
SEQUENCE OF. This is the definition of a vector, with all elements having the same data type. If
each element has a simple data type, such as an integer, then we have a simple vector (a
one-dimensional array). But we’ll see that SNMP uses this data type with each element of the
vector being a SEQUENCE (structure). We can then think of it as a two-dimensional array or
table.

For example, the UDP listener table is named udpTable and it is a SEQUENCE OF the
2-element SEQUENCE (structure) UdpEntry that we just described. Figure 25.5 shows this
two-dimensional array.

[IMAGE]
Figure 25.5 UDP listener table (udpTable) as a two-dimensional array in SNMP.

The number of rows in these tables is not specified by SNMP, but we’ll see that a manager using the
get-next operator (Section 25.7) can determine when the final row of a table has been returned.
Also, in Section 25.6 we’ll see how the manager specifies which row of a table it wants to get or set.

25.4 Object Identifiers
An object identifier is a data type specifying an authoritatively named object. By "authoritative" we
mean that these identifiers are not assigned randomly, but are allocated by some organization that has
responsibility for a group of identifiers.

348

An object identifier is a sequence of integers separated by decimal points. These integers traverse a
tree structure, similar to the DNS (Figure 14.1) or a Unix filesystem. There is an unnamed root at the
top of the tree where the object identifiers start. (This is the same direction of tree traversal that’s used
with a Unix filesystem.)

Figure 25.6 shows the structure of this tree when used with SNMP. All variables in the MIB start with
the object identifier 1.3.6.1.2.1.

Each node in the tree is also given a textual name. The name corresponding to the object identifier
1.3.6.1.2.1 is iso.org.dod.internet.mgmt.mib. These names are for human readability. The
names of the MIB variables that are in the packets exchanged between the manager and agent (Figure
25.2) are the numeric object identifiers, all of which begin with 1.3.6.1.2.1.

[IMAGE]
Figure 25.6 Object identifiers in the Management Information Base.

Besides the mib object identifiers in Figure 25.6 we also show one named
iso.org.dod.internet.private.enterprises (1.3.6.1.4.1). "This is where
vendor-specific MIBs are located. The Assigned Numbers RFC lists around 400 identifiers registered
below this node.

25.5 Introduction to the Management Information Base
The Management Information Base, or MIB, is the database of information maintained by the agent
that the manager can query or set. We describe what’s called MIB-II, specified in RFC 1213
[McCloghrie and Rose 1991].

As shown in Figure 25.6, the MIB is divided into groups named system, interfaces, at (address
translation), ip, and so on.

In this section we describe only the variables in the UDP group. This is a simple group with only a few
variables and a single table. In the next sections we use this group to show the details of instance
identification, lexicographic ordering, and some simple examples of these features. After these
examples we return to the MIB in Section 25.8 and describe some of the other groups in the MIB.

In Figure 25.6 we showed the group named udp beneath mib. Figure 25.7 shows the structure of the
UDP group.

[IMAGE]
Figure 25.7 Tree structure of IP address table.

There are four simple variables and a table containing two simple variables. Figure 25.8 describes the
four simple variables.

349

Name Datatype R/W Description
udpInDatagrams Counter Number of UDP datagrams delivered to user processes.
Number of received UDP datagrams for which no
udpNoPorts Counter
application process was at the destination port.
Number of undeliverable UDP datagrams for reasons
udpInErrors Counter other than no application at destination port (e.g., UDP
checksum error).
udpOutDatagrams Counter Number of UDP datagrams sent.

Figure 25.8 Simple variables in udp group.

We’ll use this format to describe all the MIB variables in this chapter. The column labeled "R/W" is
empty if the variable is read-only, or contains a bullet (*) if the variable is read-write. We always
include this column, even if all the variables in a group are read-only (since they are in the udp group)
to reiterate that none of the variables can be set by the manager. Also, when the data type is an
INTEGER with bounds, we specify the lower limit and upper limit, as we do for the UDP port number
in the next figure.

Figure 25.9 describes the two simple variables in the udpTable.

UDP listener table, index = < udpLocalAddress >.< udpLocalPort >
Local IP address for this listener. 0.0.0.0 indicates
udpLocalAddress lpAddress the listener is willing to receive datagrams on any
interface.
udpLocalPort [0..65535] Local port number for this listener.

Figure 25.9 Variables in udpTable.

Each time we describe the variables in an SNMP table, the first row of the figure indicates the value of
the "index" used to reference each row of the table. We show some examples of this in the next
section.

Case Diagrams
There is a relationship between the first three counters in Figure 25.8. Case Diagrams [Case and
Partridge 1989] visually illustrate the relationships between the various MIB variables in a given
group. Figure 25.10 is a Case Diagram for the UDP group.

[IMAGE]
Figure 25.10 Case Diagram for UDP group.

What this diagram shows is that the number of UDP datagrams delivered to applications
(udpInDatagrams) is the number of UDP datagrams delivered from IP to UDP, minus
udpInErrors, minus udpNoPorts. Also, the number of UDP datagrams delivered to IP
(udpOutDatagrams) is the number passed to UDP from the applications. This illustrates that

350

udpInDatagrams does not include udpInErrors or udpNoPorts.

These diagrams were used during the development of the MIB to verify that all data paths for a packet
were accounted for. [Rose 1994] shows Case Diagrams for all the groups in the MIB.

25.6 Instance Identification
Every variable in the MIB must be identified when SNMP is referencing it, to fetch or set its value.
First, only leaf nodes are referenced. SNMP does not manipulate entire rows or columns of tables.
Returning to Figure 25.7, the leaf nodes are the four that we described in Figure 25.8 and the two in
Figure 25.9. mib, udp, udpTabie, and udpEntry are not leaf nodes.

Simple Variables
Simple variables are referenced by appending ".0" to the variable’s object identifier. For example, the
counter udpInDatagrams from Figure 25.8, whose object identifier is 1.3.6.1.2.1.7.1, is referenced
as 1.3.6.1.2.1.7.1.0. The textual name of this reference is
iso.org.dod.internet.mgmt.mib.udp.udpInDatagrams.0.

Although references to this variable are normally abbreviated as just udpInDatagrams.0, we
reiterate that the name of the variable that appears in the SNMP message (Figure 25.2) is the object
identifier 1.3.6.1.2.1.7.1.0.

Tables
Instance identification of table entries is more detailed. Let’s return to the UDP listener table (Figure
25.7).

One or more indexes are specified in the MIB for each table. For the UDP listener table, the MIB
defines the index as the combination of the two variables udpLocalAddress, which is an IP
address, and udpLocalPort, which is an integer. (We showed this index in the top row in Figure
25.9.)

Assume there are three rows in the UDP listener table: the first row is for IP address 0.0.0.0 and port
67, the second for 0.0.0.0 and port 161, and the third for 0.0.0.0 and port 520. Figure 25.11 shows this
table.

udpLocalAddress udpLocalPort
0.0.0.0 67
0.0.0.0 161
0.0.0.0 520

Figure 25.11 Sample UDP listener table.

This implies that the system is willing to receive UDP datagrams on any interface for ports 67
(BOOTP server), 161 (SNMP), and 520 (RIP). The three rows in the table are referenced as shown in
Figure 25.12.

351

Lexicographic Ordering
There is an implied ordering in the MIB based on the order of the object identifiers. All the entries in
MIB tables are lexicographically ordered by their object identifiers. This means the six variables in
Figure 25.12 are ordered in the MIB as shown in Figure 25.13. Two key points result from this
lexicographic ordering.

Row Object identifier Abbreviated name Value
1.3.6.1.2.1.7.5.1.1.0.0.0.0.67 udpLocalAddress.0.0.0.0.67 0.0.0.0
1
1.3.6.1.2.1.7.5.1.2.0.0.0.0.67 udpLocalPort.0.0.0.0.67 67
2
1.3.6.1.2.1.7.5.1.2.0.0.0.0.161 udpLocalPort.0.0.0.0.161 161
3
1.3.6.1.2.1.7.5.1.2.0.0.0.0.520 udpLocalPort.0.0.0.0.520 520

Figure 25.12 Instance identification for rows in UDP listener table.

Object identifier
Column Abbreviated name Value
(lexicographically ordered)
1 1.3.6.1.2.1.7.5.1.1.0.0.0.0.161 udpLocalAddress.0.0.0.0.161 0.0.0.0
1.3.6.1.2.1.7.5.1.2.0.0.0.0.67 udpLocalPort.0.0.0.0.67 67
2 1.3.6.1.2.1.7.5.1.2.0.0.0.0.161 udpLocalPort.0.0.0.0.161 161
1.3.6.1.2.1.7.5.1.2.0.0.0.0.520 udpLocalPort.0.0.0.0.520 520

Figure 25.13 Lexicographic ordering of UDP listener table.

1. Since all instances for a given variable (udpLocalAddress) appear before all instances for the
next variable in the table (udpLocalPort), this implies that tables are accessed in a
column-row order. This results from the lexicographic ordering of the object identifiers, not the
human-readable names.
2. The ordering of the rows in a table depends on the values of the indexes for the table. In Figure
25.13, 67 is lexicographically less than 161, which is lexicographically less than 520.

Figure 25.14 shows this column-row order for our sample UDP listener table.

[IMAGE]
Figure 25.14 UDP listener table, showing column-row ordering.

We’ll also see this column-row ordering when we use the get-next operator in the next section.

352

25.7 Simple Examples
In this section we’ll show some examples that fetch the values of variables from an SNMP agent. The
software used to query the agent is called snmpi and is from the ISODE system. Both are described
briefly in [Rose 1994].

Simple Variables
We’ll query a router for two simple variables from the UDP group:

sun % snmpi -a gateway -c secret

snmpi> get udpInDatagrams.0 udpNoPorta.0
udplnDatagrams.0=616168
udpNoPorts.0=33

snmpi> quit

The -a option identifies the agent we want to communicate with, and the -c option specifies the
SNMP community. It is a password supplied by the client (snmpi in this case) and if the server (the
agent in the system gateway) recognizes the community name, it honors the manager’s request. An
agent could allow clients within one community read-only access to its variables, and clients in
another community read-write access.

The program outputs its snmpi> prompt, and we can type commands such as get, which translates
into an SNMP get-request message. When we’re done, we type quit. (In all further examples
we’ll remove this final quit command.) Figure 25.15 shows the two lines of tcpdump output for
this example.

sun.1024 > gateway.161: GetRequest(42)
1 0.0
1.3.6.1.2.1.7.1.0 1.3.6.1.2.1.7.2.0
gateway.161 > sun.1024: GetResponse(46)
2 0.348875 (0.3489) 1.3.6.1.2.1.7.1.0=616168
1.3.6.1.2.1.7.2.0=33

Figure 25.15 tcpdump output for simple SNMP query.

The request for the two variables is sent in a single UDP datagram, and the response is also a single
UDP datagram.

We show the variables as their respective object identifiers, because that is what’s sent in the SNMP
messages. We had to specify the instance of the two variables as 0. Notice also that the name of the
variable (its object identifier) is always returned in the response. We’ll see below that this is required
for the get-next operator to work.

get-next Operator
The operation of the get-next operator is based on the lexicographic ordering of the MIB. We start
the following example by asking for the next object identifier after udp (without specifying any
instance, since this is not a leaf object). This returns the first object in the UDP group. We then ask for

353

the next entry after this one, and the second entry is returned. We repeat this one more time to get the
third entry:

sun % sninpi -a gateway -c secret

snmpi> next udp
udpInDatagrams.0=616318

snmpi> next udpInDatagraros.0
udpNoPorts.0=33

snmpi> next udpNoPorts.0
udpInErrors.0=0

This example shows why a get-next operator must return the name of the variable: we ask the
agent for the next variable, and the agent returns its name and value.

Using the get-next operator in this fashion, one could imagine a manager with a loop that starts at
the beginning of the MIB and queries the agent for every variable that the agent maintains. Another
use of this operator is to iterate through tables.

Table Access
We can reiterate the column-row ordering of tables using our simple query program to step through the
entire UDP listener table. We start by asking for the next variable after udpTable. Since this is not a
leaf object we can’t specify an instance, but the get-next operator still returns the next object in the
table. We then work our way through the table, with the agent returning the next variable, in
column-row order:

sun % snmpi-a gateway -c secret

snmpi> next udpTable
udpLocalAddress.0.0.0.0.67=0.0.0.0

snmpi> next udpLocalAddress.0.0.0.0.67
udpLocalAddress.0.0.0.0.161-0.0.0.0

udpLocalAddress.0.0.0.0.52 0=0.0.0.0

udpLocalPort.0.0.0.0.67=67

snmpi> next udpLocalPort.0.0.0.0.67


snmpInPkts.0=59 we’re finished with the UDP listener table

354

We see that the order returned corresponds to Figure 25.14.

How does a manager know when it reaches the end of a table? Since the response to the get-next
operator contains the name of the next entry in the MIB after the table, the manager can tell when the
name changes. In our example the last entry in the UDP listener table is followed by the variable
snmpInPkts.

25.8 Management Information Base (Continued)
We now return to the description of the MIB. We describe only the following groups: system
(system identification), if (interfaces) , at (address translation), ip, icmp, and tcp. Additional
groups are defined.

system Group
The system group is simple; it consists of seven simple variables (i.e., no tables). Figure 25.16
shows their names, data types, and descriptions.

Display
sysDescr Textual description of entity.
String
sysObjectID ObjectID Vendor’s ID within the subtree 1.3.6.1.4.1.
Time in hundredths of a second since network
sysUpTime TimeTicks
management portion of system was rebooted.
sysContact DisplayString * Name of contact person and how to contact them.
sysName DisplayString * Node’s fully qualified domain name (FQDN).
sysLocation DisplayString * Physical location of node.
Value indicating services provided by node. It is the
sum of the layers in the OSI model supported by the
node. The following values are added together,
sysServices [0..127]
depending on the services provided: 0x01 (physical) ,
0x02 (datalink), 0x04 (internet), 0x08 (end-to-end),
0x40 (application).

Figure 25.16 Simple variables in system group.

We can query the router netb for some of these variables:

sun % snmpi -a netb -c secret

snmpi> get sysDescr.0 sysObjectID.0 sysDpTime.0 sysServices.0
sysDescr.0="Epilogue Technology SNMP agent for Telebit NetBlazer"
sysObjectID.0=1.3.6.1.4.1.12.42.3.1
sysUpTime.0=22 days, 11 hours, 23 minutes, 2 seconds (194178200
timeticks)
sysServices.0=0xc<internet,transport>

355

The system’s object identifier is in the internet.private.enterprises group (1.3.6.1.4.1)
from Figure 25.6. From the Assigned Numbers RFC we can determine that the next object identifier
(12) is assigned to the vendor (Epilogue).

We can also see that the sysServices variable is the sum of 4 and 8: this element supports the
Internet layer (i.e., routing) and the transport layer (i.e., end-to-end).

interface Group
Only one simple variable is defined for this group: the number of interfaces on the system, shown in
Figure 25.17.

ifNumber INTEGER Number of network interfaces on system.

Figure 25.17 Simple variable in if group.

This group also defines a table with 22 columns. Each row of the table defines the characteristics for
each interface, as shown in Figure 25.18.

Interface table, index = < IfIndex >
Index of interface, between one and if
ifIndex INTEGER
Number.
ifDescr DisplayString Textual description of interface.
Type, for example: 6 = Ethernet, 7 = 802.3
ifType INTEGER Ethernet, 9 = 802.5 token ring, 23 = PPP, 28
= SLIP, and many other values.
ifMtu INTEGER MTU of interface.
ifSpeed Gauge Speed in bits/sec.
Physical address, or string of 0 length for
ifPhysAddress PhysAddress interfaces without physical addresses (e.g.,
serial links).
Desired state of interface: 1 = up, 2 = down, 3
ifAdminStatus [1..3] *
= testing.
Current state of interface: 1 = up, 2 = down,
ifOperStatus [1..3]
3 = testing.
Value of sysUpTime when interface
ifLastChange TimeTicks
entered current operational state.
Total number of bytes received, including
ifInoctets Counter
framing characters.

356

Number of unicast packets delivered to
ifInUcastPkts Counter
higher layers.
Number of nonunicast (i.e., broadcast or
ifInNUcastPkts Counter
multicast) packets delivered to higher layers.
Number of received packets discarded even
ifInDiscards Counter
though no error in packet (i.e., out of buffers).
Number of received packets discarded
ifInErrors Counter
because of errors.
Number of received packets discarded
ifInUnknownProtos Counter
because of unknown protocol.
Number of bytes transmitted, including
ifOutoctets Counter
framing characters.
Number of unicast packets received from
ifOutUcastPkts Counter
higher layers.
Number of nonunicast (i.e., broadcast or
ifOutNUcastPkts Counter multicast) packets received from higher
layers.
Number of outbound packets discarded even
ifOutDiscards Counter
though no error in packet (i.e., out of buffers).
Number of outbound packets discarded
ifOutErrors Counter
because of errors.
ifOutQLen Gauge Number of packets in output queue.
A reference to MIB definitions specific to
ifSpecific ObjectID
this particular type of media.

Figure 25.18 Variables in interface table: ifTable.

We can query the host sun for some of these variables for all its interfaces. Wet expect to find three
interfaces, as in Section 3.8, if the SLIP interface is up:

sun % snmpi -a sun

snropi> next ifTable first see what index of first interface is
if Index.1=1

snmpi> get ifDescr.1 if Type.1 ifMtu.1 if Speed.1 ifPhysAddress.1
ifDescr.1="le0"
if Type.1=ethernet-csmacd(6)
ifMtu.1=1500
ifSpeed, 1=10000000
ifPhysAddress.1=0x08:00:20;03:f6:42

357

snmpi> next ifDescr.1 if Type.1 ifMtu.1 if Speed.1 ifPhysAddress.1

ifDescr.2="sl0"
ifType.2=propPointToPointSerial(22)
ifMtu.2=552
ifSpeed.2=0
ifPhysAddress.2=0x00:00:00:00:00:00

snmpi> next ifDescr.2 ifType.2 ifMtu.2 if Speed.2 ifPhysAddress.2
ifDescr.3="lo0"
ifType.3=softwareLoopback(24)
ifMtu.3=1536
ifSpeed.3=0
ifPhysAddress.3=0x00:00:00:00:00:00

We first get five variables for the first interface using the get operator, and then get the same five
variables for the second interface using the get-next operator. The last command gets these same
five variables for the third interface, again using the get-next command.

The interface type for the SLIP link is reported as proprietary point-to-point serial, not SLIP. Also, the
speed of the SLIP link is not reported.

It is critical to understand the relationship between the get-next operator and the column-row
ordering. When we say next ifDescr.1 it returns the next row of the table for this variable, not the
next variable in the same row. If tables were stored in a row-column order instead, we wouldn’t be
able to step to the next occurrence of a given variable this way.

at Group
The address translation group is mandatory for all systems, but was deprecated by MIB-II. Starting
with MIB-II, each network protocol group (e.g., IP) contains its own address translation tables. For IP
it is the ipNetToMediaTable.

Only a single table with three columns is defined for the at group, shown in Figure 25.19.

We can use a new command within the snmpi program to dump an entire table. We’ll query the
router named kinetics (which routes between a TCP/IP network and an AppleTalk network) for its
entire ARP cache. This output reiterates the lexicographic ordering of the entries in the table:

Address translation table, index = < atIflndex >.1.< atNetAddress >
atiflndex INTEGER * Interface number: ifIndex.
Physical address. Setting this to a string of 0
atPhysAddress PhysAddress *
length invalidates the entry.
atNetAddress NetworkAddress * IP address.

358

Figure 25.19 Address translation table: atTable.

sun % snmpi -a kinetics -c secret dump at

atiflndex.1.1.140.252.1.4=1
atiflndex.1.1.140.252.1.22=1
atiflndex.1.1.140.252.1.183=1
atiflndex.2.1.140.252.6.4=2
atiflndex.2.1.140.252.6.6=2

atPhysAddress.1.1.140.252.1.4=0xaa:00:04:00:f4:14
atPhysAddress.1.1.140.252.1.22=0x08:00:20:0f:2d:38
atPhysAddress.1.1.140.252.1.183=0x00:80:ad:03:6a:80
atPhysAddress.2.1.140.252.6.4=0x00:02:16:48
atPhysAddress.2.1.140.252.6.6=0x00:02:3c:48

atNetAddress.1.1.140.252.1.4=140.252.1.4
atNetAddress.1.1.140.252.1.22=140.252.1.22
atNetAddress.1.1.140.252.1.183=140.252.1.183
atNetAddress.2.1.140.252.6.4=140.252.6.4
atNetAddress.2.1.140.252.6.6=140.252.6.6

If we watch the packet exchange using tcpdump, when snmpi dumps an entire table it first issues a
get-next for the table name (at in this example) to get the first entry. It prints the first entry and
issues another get-next. This continues until the entire table has been dumped.

Figure 25.20 shows the arrangement of this table.

atIflndex atPhysAddress atNetAddress
1 0xaa:00:04:00:f4:14 140.252.1.4
1 0x08:00:20:0f:2d:38 140.252.1.22
1 0x00:80:ad:03:6a:80 140.252.1.183
2 0x00:02:16:48 140.252.6.4
2 0x00:02:3c:48 140.252.6.6

Figure 25.20 Example of at table (ARP cache).

The AppleTalk physical addresses on interface number 2 are 32-bit values, not the 48-bit Ethernet
addresses to which we’re accustomed. Also note that an entry exists for our router (netb at
140.252.1.183), which we expect, since kinetics and netb are on the same Ethernet (140.252.1)
and kinetics must use ARP to send the SNMP responses back to us.

ip Group
The ip group defines numerous variables and three tables. Figure 25.21 defines the simple variables.

359

1 means the system is forwarding IP datagrams, and 2
ipForwarding [1..2] *
means it is not.
Default TTL value when transport layer doesn’t
ipDefaultTTL INTEGER *
provide one.
Total number of received IP datagrams from all
ipInReceives Counter
interfaces.
Number of IP datagrams discarded because of header
ipInHdrErrors Counter errors (e.g., checksum error, version number
mismatch, TTL exceeded, etc.).
Number of IP datagrams discarded because of
ipInAddrErrors Counter
incorrect destination address.
Number of IP datagrams for which an attempt was
ipForwDatagrams Counter
made to forward.
Number of locally addressed IP datagrams with an
ipInUnknownProtos Counter
invalid protocol field.
Number of received IP datagrams discarded because
ipInDiscards Counter
of a lack of buffer space.
Number of IP datagrams delivered to appropriate
ipInDelivers Counter
protocol module.
Total number of IP datagrams passed to IP for
ipOutRequests Counter transmission. Does not include those counted in
ipForwDatagrams.
Number of output IP datagrams discarded because of a
ipOutDiscards Counter
lack of buffer space.
Number of IP datagrams discarded because no route
ipOutNoRoutes Counter
could be found.
Maximum number of seconds that received fragments
ipReasmTimeout INTEGER
are held while awaiting reassembly.
Number of IP fragments received that needed to be
ipReasmReqds Counter
reassembled.
ipReasmOKs Counter Number of IP datagrams successfully reassembled.
ipReasmFails Counter Number of failures by IP reassembly algorithm.
Number of IP datagrams that have been successfully
ipFragOKs Counter
fragmented.
Number of IP datagrams that needed to be fragmented
ipFragFails Counter
but couldn’t because the "don’t fragment" flag was set.

360

ipFragCreates Counter Number of IP fragments generated by fragmentation.
Number of routing entries chosen to be discarded even
ipRoutingDiscards Counter
though they were valid.

Figure 25.21 Simple variables in ip group.

The first table in the ip group is the IP address table. It contains one row for each IP address on the
system. Each row contains five variables, described in Figure 25.22.

IP address table, index = < ipAdEntAddr >
ipAdEntAddr lpAddress IP address for this row.
ipAdEntIf Index INTEGER Corresponding interface number: ifIndex.
ipAdEntNetMask lpAddress Subnet mask for this IP address.

Libro ilustrate dstevens tcpip

More Related Content

What's hot (13)

Viewers also liked (16)

Similar to Libro ilustrate dstevens tcpip (20)

Libro ilustrate dstevens tcpip