5 data link-lan.key

Computer Networking :
Principles, Protocols and Practice
Part 5
Datalink Layer and Local Area Network

Olivier Bonaventure
http://inl.info.ucl.ac.be/

CNP3/2008.5. © O. Bonaventure 2008

Datalink layer

 Point-to datalink layer
 How to transmit and receive frames

 Local area networks
 Optimistic Medium access control
 ALOHA, CSMA, CSMA/CD, CSMA/CA

 Ethernet networks
 WiFi networks
 Deterministic Medium access control
 Token Ring, FDDI

CNP3/2008.5. © O. Bonaventure, 2008

Usage of the physical layer

 Service provided by physical layer
 Bit transmission between nodes attached to the
same physical transmission channel
 cable, radio, optical ﬁber, ...

 Better service for computers
 Transmission/reception of short messages
 Service provided by the datalink layer

2 Datalink Datalink

1 Physical Physical


Datalink layer

 Goals
 Provide a reliable transfert of packets although
 Physical layer sends/receives bits and not packets
 Physical layer service is imperfect
 transmission errors
 Losses of bits
 Creation of bits

Packet

Datalink Datalink


Datalink layer

 Goals
 Provide a reliable transfert of packets although
 Physical layer sends/receives bits and not packets
 Physical layer service is imperfect
 transmission errors
 Losses of bits
 Creation of bits

Packet Packet

Datalink Datalink


Frame delineation
 Frame
 Unit of information transfer between two entities of
the datalink layer
 sequence of N bits
 Datalink layer usually supports variable-length frames

Packet Packet

Datalink Frames Datalink

Physical Bits Physical

 How can the receiver extract the frames from the
received bit stream ?


Frame delineation

 Naïve solutions
 Use frame size to delineate frames
 Insert frame size in frame header
 Issue
 What happens when errors affect frame payload and frame
header ?

 Use special character/bitstring to mark beginning/
end of frame
 Example
 all frames start with #
 Issue
 What happens when the special character/bitstring appears
inside the frame payload ?


Character stufﬁng

 Character stufﬁng
 Suitable for frames containing an integer number
of bytes
 'DLE' 'STX' to indicate beginning of frame
 'DLE' 'ETX' to indicate end of frame
 When transmitting frame, sender replaces 'DLE' by
'DLE' 'DLE' if 'DLE' appears inside the frame
 Receiver removes 'DLE' if followed by 'DLE'

 Example
 Packet : 1 2 3 'DLE' 4
 Frame
'DLE' 'STX' 1 2 3 'DLE' 'DLE' 4 'DLE' 'ETX'


Bit stuffing

 Alternative to character stuffing
 Suitable for frames composed of n bits
 01111110 used as marker at beginning and end of
frame
 Sender behaviour
 If five bits set to ʻ1ʼ must be sent, sender adds a bit set to
ʻ0ʼ immediately after the fifth bit set to ʻ1ʼ
 Receiver behaviour
 Counts the number of successive bits set to 1
 6 successive bits set to 1 followed by 0 : marker
 5 successive bits set to 1 followed by 0 : remove bit set to 0

 Example
 Packet : 0110111111111111111110010
 Frame
01111110011011111011111011111011001001111110


Frame delineation

 Co-operation with physical layer
 Some physical layers are able to transmit special
physical codes that represent neither 0 nor 1
 Example : Manchester coding

1 0 1 0 0 1 1 0
InvHInvB

 invH (or N times invH) could be used to mark the beginning
of a frame and invB (or N times invB) to mark the end of a
frame


Frame delineation in practice

 Most datalink protocols use
 Character stuffing or bit stuffing
 Character stuffing is preferred by software
implementations
 A length field in the frame header
 A checksum or CRC in the header or trailer to
detect transmission errors

 A receiver frame is considered valid if
 the correct delimiter appears at the beginning
 the length is correct
 the CRC/checksum is valid
 the correct delimiter appears at the beginning


PPP : Point-to-Point Protocol

 Goal
 Allow the transmission of network layer (IP but
also other protocols) packets over serial lines
 modems, leased lines, ISDN, ...
 Architecture
 PPP is composed of three different protocols
1. PPP
 transmission of data frames (e.g. IP packets)
2. LCP : Link Control Protocol
• Negotiation of some options and authentication
(username, password) and end of connection
3. NCP : Network Control Protocol
• Negotiation of options related to the network layer
protocol used above PPP
(ex: IP address, IP address of DNS resolver, ...)


PPP (2)

 PPP frame format
Flag Address Control Checksum Flag
01111110 11111111 00000011 Protocol Packet 16bits/32bits 01111110

Identification of the network layer packet
transported in the PPP frame
 Mechanisms used by PPP
 character stuffing for asynchronous lines
 bit stuffing for synchronous lines
 CRC for error detection
 16 bits default but 32 bits CRC can be negotiated
 No error correction by default
 a reliable protocol can be negotiated
 Data compression option
 content of PPP frames can be compressed. To be negotiated at
beginning of PPP connection


DataLink layer

 How to transmit and receive frames


 WiFi networks
 Token Ring, FDDI


Local area networks
 How to efﬁciently connect N hosts together ?
 Ideally we would like to have a single cable on
each host while being able to reach all the others

 Network topologies
 Star-shaped network
 Ring-shaped network
 Bus-shaped network


Local area networks
 Problems to be solved
 How to identify the hosts attached to the LAN ?
 The LAN is a shared resource
 How to regulate access to this shared resource to
provide :
 fairness
 All hosts should be able to use a fair fraction of the shared
resource
 performance
 The shared resource should be used efﬁciently


Static allocation

 Time Division Multiplexing

A D

B C

T
A B C DA B C D A B C D A B C D A B C D A B C D A B C D
DA


Static allocation

 Time Division Multiplexing

A D

B C

T
A B C DA B C D A B C D A B C D A B C D A B C D A B C D
DA

 No suitable for a computer network
 Leads to low link utilisation and high delays
 Computers generate bursty traﬁc
 A more adaptive access control mechanism is
required

Medium access control

 Hypotheses
 N stations need to share the same transmission
channel
 A single transmission channel is available
 Deﬁnition
 Collision
 If two stations transmit their frame at the same time, their
electrical signal appears on the channel and causes a collision

 Options
 Frame transmission
 A station can transmit at any time
 A station can only transmit at speciﬁc instants
 Listening while transmitting
 A station can listen while transmitting
 A station cannot listen while transmitting


Medium access control
 How to regulate access to the shared
medium ?
 Statistical or optimistic solutions
 hosts can transmit frames at almost any time
 if the low is low, the frames will arrive correctly at destination
 if the low is high, frames may collide
 distributed algorithm allows to recover from the
collisions
 Deterministic or pessimistic solutions
 Collisions are expensive and need to be avoided
Distributed algorithm distributes authorisations to
transmit to ensure that a single host is allowed to
transmit at any time
 avoids collisions when load is high, but may delay transmission
when load is low


ALOHA

 Problem
 terminals need to exchange data with computer
but phone lines are costly

 Solution
 Wireless network
 upstream frequency shared by all terminals
 terminals do not hear each other
 downlink frequency reserved for computer

ALOHA (3)
 How to organise frame transmission ?
 If a host is alone, no problem
 If two hosts transmit at the same time, a collision
will occur and it will be impossible to decode their
transmission

A
B


ALOHA (3)
 How to organise frame transmission ?
 If a host is alone, no problem
 If two hosts transmit at the same time, a collision
will occur and it will be impossible to decode their
transmission

A
B

collision

ALOHA (3)

 Medium access algorithm
 First solution

N=1;
while ( N<= max) do
send frame;
wait for ack on return channel or timeout:
if ack on return channel
exit while;
else
/* timeout */
/* retransmission is needed */
N=N+1;
end do
/* too many attempts */


ALOHA (4)

 Drawback
 When two stations enter in collision, they may
continue to collide after

T0 T3 T3+timeout

A
B Collision

T1 T2 T2+timeout

 How to avoid this synchronisation among stations ?


ALOHA (5)

 Improved algorithm


N=1;
while ( N<= max) do
send frame;
wait for ack on return channel or timeout:
if ack on return channel
exit while;
else
/* timeout */
wait for random time;
N=N+1;
end do


Slotted ALOHA

 How to improve ALOHA
 Idea
 Divide time in slots and force stations to only transmit
frames at the beginning of slots
 requires a common clock heard by all stations
 decrease risk of collisions since a slot is either
 empty (no frame)
 full (a single frame was transmitted)
 in collision (more than one frame was transmitted)


Slotted ALOHA

 How to improve ALOHA
 Idea
 Divide time in slots and force stations to only transmit
frames at the beginning of slots
 requires a common clock heard by all stations
 decrease risk of collisions since a slot is either
 empty (no frame)
 full (a single frame was transmitted)
 in collision (more than one frame was transmitted)

A
B

collision


Carrier Sense Multiple Access

 How to improve slotted Aloha ?
 Idea
 Stations should be polite
 Listens to the transmission channel before transmitting
 Wait until the channel becomes free to transmit

 Limitations
 Politeness is only possible if all stations can listen to
the transmission of all stations
 true when all stations are attached to the same cable, but not in
wireless networks


CSMA

 CSMA
 Carrier Sense Multiple Access

N=1;
while ( N<= max) do
wait until channel becomes free;
send frame immediately;
wait for ack or timeout:
if ack received
exit while;
else
/* timeout */
N=N+1;
end do


non-persistent CSMA
 Idea
 Transmitting a frame immediately after the end of
the previous one is a very aggressive behaviour
 If the channel is free, transmit
 Otherwise wait some random time before listening again
N=1;
while ( N<= max) do
listen channel;
if channel is empty
send frame;
wait for ack or timeout
if ack received
exit while;
else /* retransmission is needed */
N=N+1:
else
end do


p-persistent CSMA
 Tradeoff between CSMA and non-persistent
CSMA

N=1;
while ( N<= max) do
listen channel;
if channel is empty
with probability p
send frame;
wait for ack or timeout
if ack received
exit while;
else /* retransmission needed */
N=N+1;
else
end do


Improvements to CSMA

 Problems with CSMA
 If one bit of a frame is affected by a collision, the
entire frame is lost

 Solution
 Stop the transmission of a frame as soon as a
collision has been detected

 How to detect collisions ?
 Station listens to channel while transmitting
 If there is no collision, it will hear the signal it transmits
 If there is a collision, is will hear an incorrect signal

 CSMA/CD
 Carrier Sense Multiple Access with Collision
CNP3/2008.5.
Detection © O. Bonaventure, 2008

CSMA/CD

 Medium access control
N=1;
while ( N<= max) do
send frame and listen;
wait until (end of frame) or (collision)
if collision detected
stop transmitting;
/* after a special jam signal */
else
/* no collision detected */
wait for interframe delay;
exit while;
N=N+1;
end do


CSMA/CD : Example

A B


CSMA/CD : Example

A B

1
Start of frame


CSMA/CD : Example

A B

1
Start of frame

2
Frame is propagated on LAN (5 microsecond per kilometer)


CSMA/CD : Example

A B

1
Start of frame

2

3

Frame stops at left side and ﬁrst bit reaches B


CSMA/CD : Example

A B

1
Start of frame

2

3

Frame stops at left side and ﬁrst bit reaches B

4
Frame leaves the LAN

CSMA/CD : Collisions

A B



A B

1
Frame starts at A and B almost at the same time



A B

1
Collision : at this point on the
shared medium, it is impossible
to decode the signal

2



A B

1
Collision : at this point on the
shared medium, it is impossible
to decode the signal

2

A detects the collision and
stops transmitting its frame

3


CSMA/CD : Collisions (2)
 Advantages
 Improves channel utilisation as stations do not
transmit corrupted frames
 a station can detect whether its frame was sent
without collision
 implicit acknowledgement if destination is up
 when a collision is detected, automatic retransmission

 Is it possible for a station to detect all
collisions on all its frames ?



A B



A B

1
A and B send a small frame almost at the same time



A B

1

2
The two frame collide in the middle of the medium



A B

1

2
The two frame collide in the middle of the medium

3
A did not notice any collision
B did not notice any collision
They both consider that their frames were received correctly


 How to ensure that all collisions are detected ?
 Worst case scenario

A B



A B

1 Start of the frame sent by A



A B


2
After τ seconds, Aʼs frame reaches B
A time τ−ε, B starts to transmit its own frame
B notices the collision immediately and stops transmitting



A B


2
After τ seconds, Aʼs frame reaches B
A time τ−ε, B starts to transmit its own frame
B notices the collision immediately and stops transmitting

3
A detects collision at time τ+τ−ε


 How can a station ensure that it will be able
to detect all the collisions affecting its
frames ?
 Each frame must be transmitted for at least a
duration equal to the two way delay (2∗τ)
 As the throughput on a bus is ﬁxed, if the two way delay
is ﬁxed, then all frames must be larger than a minimum
frame size
 Improvement
 To ensure that all stations detect collisions, a station that notices
a collision should send a jamming signal


Exponential backoff

 How to deal with collisions ?
 If the stations that collide retransmit together, a
new collision will happen

 Solution
 Wait some random time after the collision
 After collision, time is divided in slots
 a slot = time required to send a minimum sized frame
 After first collision, wait 0 or 1 slot before retransmitting
 After first collision, wait 0, 1,2 or 3 slots before retransmitting
 After first collision, wait 0..2i-1 slots before retransmitting


CSMA/CD with exponential backoff

N=1;
while ( N<= max) do
send frame and listen;
wait until (end of frame) or (collision)
if collision detected
stop transmitting;
/* after a special jam signal */
k = min (10, N);
r = random(0, 2k - 1) * slotTime;
wait for r time slots;
else
/* no collision detected */
wait for interframe delay;
exit while;
N=N+1;
end do


CSMA with Collision Avoidance

 Goal
 Design a medium access control method suitable for
wireless networks
 on a wireless network, a sender cannot usually listen to its
transmission (and thus CSMA/CD cannot be used)

 Improvements to CSMA
 Initial delay before transmitting if channel is empty
 Extended Inter Frame Space (EIFS)
 Minimum delay between two successive frames
 Distributed Coordination Function Inter Frame Space (DIFS)
 Delay between frame reception and ack transmission
 Short Inter Frame Spacing (SIFS, SIFS< DIFS < EIFS)


CSMA/CA (1)

 Sender
N=1;
while ( N<= max) do
if (previous frame corrupted)
{ wait until channel free during t>=EIFS; }
else
{ wait until endofframe;
wait until channel free during t>=DIFS; }
send data frame ;
if ack received
exit while;
else
/* timeout retransmission is needed */
N=N+1;
end do


CSMA/CA (2)

 Receiver

While (true)
{
Wait for data frame;
if not(duplicate)
{ deliver (frame) }
wait during SIFS;
send ack (frame) ;
}


CSMA/CA : Example

B C
A


CSMA/CA : Example

B C
A
DIFS


CSMA/CA : Example

B C
A
DIFS

Data frame


CSMA/CA : Example

B C
A
DIFS

Data frame

SIFS


CSMA/CA : Example

B C
A
DIFS

Data frame

SIFS
ACK frame


CSMA/CA : Example

B C
A
DIFS

Data frame
Busy

SIFS
ACK frame


CSMA/CA : Example

B C
A
DIFS

Data frame
Busy

SIFS
ACK frame

DIFS

Data frame


CSMA/CA : Problem

B C
A


CSMA/CA : Problem

B C
A
DIFS
DIFS


CSMA/CA : Problem

B C
A
DIFS
DIFS

Busy Data frame Data frame
Busy


CSMA/CA : Problem

B C
A
DIFS
DIFS

Busy

Timeout Timeout


CSMA/CA : Problem

B C
A
DIFS
DIFS

Busy

Timeout Timeout

Data frame
Data frame
Busy


CSMA/CA
First improvement (2)
 Sender
N=1;
while ( N<= max) do
if (previous frame corruped)
{ wait until channel free during t>=EIFS; }
else
{ wait until endofframe;
wait until channel free during t>=DIFS; }
backoff_time = int(random[0,min(255,7*2N-1)])*T
wait(backoff_time)
if (channel still free)
{ send data frame ;
if ack received
exit while;
else /* timeout retransmission is needed */
N=N+1; }
end do


CSMA/CA : Example 2
A B C


CSMA/CA : Example 2
A B C

DIFS
DIFS


CSMA/CA : Example 2
A B C

DIFS
DIFS
Backoff[0,7]
Backoff[0,7]


CSMA/CA : Example 2
A B C

DIFS
DIFS
Backoff[0,7]
Backoff[0,7]
Data frame

SIFS

Ack frame


CSMA/CA : Example 2
A B C

DIFS
DIFS
Backoff[0,7]
Backoff[0,7]
Data frame
Channel busy!

SIFS Busy

Ack frame

Data frame

CSMA/CA : Example 2
A B C

DIFS
DIFS
Backoff[0,7]
Backoff[0,7]
Data frame
Channel busy!

SIFS Busy

Ack frame

End of
backoff

Data frame

CSMA/CA
Hidden station problem
 Often occurs in wireless networks

B
Hears A and C

C
A

Only hears B, not C Hears B, but not A


CSMA/CA
Second improvement
 Principle
 Allow the sender to “reserve” some air time
 Special (short) RTS frame indicates duration
 Using a short RTS frame reduces the risk of collisions while
transmitting this frame

 Allow the receiver to conﬁrm the reservation
 Special (short) CTS frame indicates reservation
 Using a short CTS frame reduces the risk of collisions while
transmitting this frame
 The stations that could collide with the
transmission will hear at least CTS
 Frame contains an indication of transmission time


CSMA/CA : Example 3
A B C


CSMA/CA : Example 3
A B C

DIFS+Backoff


CSMA/CA : Example 3
A B C

DIFS+Backoff

RTS [100 microsec]


CSMA/CA : Example 3
A B C

DIFS+Backoff

RTS [100 microsec]

SIFS
CTS[100microsec] Busy[
100microsec+
SIFS+
CTS+
SIFS+
ACK
]


CSMA/CA : Example 3
A B C

DIFS+Backoff

RTS [100 microsec]

SIFS
CTS[100microsec] Busy[
100microsec+
SIFS SIFS+
CTS+
Data [100 microsec] SIFS+
ACK
]
SIFS
ACK frame


Datalink layer


 Basics of Ethernet
 IP over Ethernet
 Interconnection of Ethernet networks

 WiFi networks
CNP3/2008.5.  Token Ring, FDDI © O. Bonaventure, 2008

Ethernet/802.3

 Most widely used LAN
 First developed by Digital, Intel and Xerox
 Standardised by IEEE and ISO

 Medium Access Control
 CSMA/CD with exponential backoff
 Characteristics
 Bandwidth: 10 Mbps
 Two ways delay
 51.2 microsec on Ethernet/802.3
 => minimum frame size : 512 bits
 Cabling
 10Base5 : (thick) coaxial cable maximum 500 m,100 stations
 10Base2 : (thin) coaxial 200 m maximum and 30 stations


Ethernet/802.3

A B C D

 Initial conﬁguration
 bus-shaped network

 Remaining problems besides CSMA/CD
 What is an Ethernet frame ?
 How does station A sends a frame to station B ?
 How does station B detects a frame
 How to support broadcast ?
 How to support multicast ?


Ethernet Addresses (2)

 Each Ethernet adapter has a unique
Ethernet address
 ensures that two hosts on the same LAN will not
use the same Ethernet address
 Ethernet Addressing format
 48 bits addresses
 Source Address
00

24 bits OUI 24 bits
(adapter manufacturer) (identiﬁer of adapter)

 Destination address
 If high order bit is 0, host unicast address
 If high order bit is 1, host multicast address
 broadcast address = 111111..111


LAN-level multicast

 Principle
 Two types of destination Ethernet addresses
 Physical addresses
 identifies one Ethernet adapter
 Logical addresses
 identifies a logical group of Ethernet destinations
48 bits
OUI
0 : physical address (unicast)
1 : logical address (multicast)

 Transmission of multicast frame
 sender transmits frame with multicast destination addr.
 Reception of multicast frames
 Ethernet adapters can be configured to capture frames
whose destination address is
 Their unicast address
 One of a set of multicast addresses

Ethernet Frames

 DIX Format
 proposed by Digital, Intel and Xerox

Preamble
[8 bytes]

Destination
address

Source address

Type
[2 bytes]

Data
[46-1500 bytes

CRC [32 bits]


Ethernet Frames

 DIX Format

Preamble Used to mark the beginning of the frame
[8 bytes] Allows the receiver to synchronise its
clock to the senderʼs clock
Destination
address

Source address

Type
[2 bytes]

Data
[46-1500 bytes

CRC [32 bits]


Ethernet Frames

 DIX Format

Destination
address

Source address

Type Indication of the type of packet contained
[2 bytes] inside the frame

Data
[46-1500 bytes

CRC [32 bits]


Ethernet Frames

 DIX Format

Destination
address

Source address

Type Indication of the type of packet contained
[2 bytes] inside the frame

Data Upper layer protocol must ensure that
[46-1500 bytes the payload of the Ethernet frame is
at least 46 bytes and at most 1500 bytes

CRC [32 bits]


802.3 Frames

Preamble
[7 bytes]
Delimiter[1byte]

Destination
Address
Source
Address

Length
[2 bytes]

Data
and padding
[ 46- 1500 bytes]

CRC [32 bits]


802.3 Frames

Delimiter[1byte] clock to the senderʼs clock
Destination
Address
Source
Address

Length
[2 bytes]

Data
and padding
[ 46- 1500 bytes]

CRC [32 bits]


802.3 Frames

 Ethernet 802.3
 standardised by IEEE

Destination
Address
Source
Address

Length
[2 bytes]

Data
and padding
[ 46- 1500 bytes]

CRC [32 bits]


802.3 Frames

 Ethernet 802.3
 standardised by IEEE

Destination
Address
Source
Address
Provides the real length of the network
Length layer packet placed inside the payload.
[2 bytes] 802.3 adds padding (bytes set to 0) to
ensure that the data ﬁeld of the frame
Data contains at least 46 bytes
and padding
[ 46- 1500 bytes]

CRC [32 bits]


Ethernet and 802.3 : details
 How can the receiver identify the type of
network protocol packet inside the frame ?
 Ethernet : thanks to Type ﬁeld
 802.3 : no Type ﬁeld !

 IEEE standard
 Divide datalink layer in two sublayers
 Medium Access Control (MAC)
 lower sublayer responsible for the frame transmission and
medium access control (CSMA/CD)
 interacts with but does not depend from the physical layer
 example : 802.3
 Logical Link Control (LLC)
 higher sublayer responsible for the exchange of frames with the
higher layers
 interacts with the higher layer
 does not depend from the MAC layer
 several variants of LLC exist

802.2 : LLC

 LLC Type 1
 Unreliable connectionless service
 Addition to 802.3
 New LLC header allows to identify upper layer protocol
 similar to Type ﬁeld of Ethernet DIX

Packet

LLC Packet LLC
P DA SA L LLC Packet CRC MAC

 LLC Type 2
 Reliable transmission with acknowledgements
 An example of a protocol developed by a
standardisation body but used by nobody...

Ethernet Service

 An Ethernet network provides a
connectionless unreliable service
 Transmission modes
 unicast
 multicast
 broadcast

 Even if in theory the Ethernet service is
unreliable, a good Ethernet network should
 deliver frames to their destination with a very hig
probability of delivery
 not reorder the transmitted frames
 reordering is obviously impossible on a bus


IPover PPP and
IP over Ethernet
IP: 10.0.0.1 Ethernet
PPP
IP: 10.0.1.9
IP: 10.0.0.2 R IP: 10.0.1.1
Eth : B
Eth : A

IP IP
PPP PPP Eth Eth
Modem Modem 10base5 10base5


IPover PPP and
IP over Ethernet
PPP
IP: 10.0.1.9
IP: 10.0.0.2 R IP: 10.0.1.1
Eth : B
Eth : A

IP IP
PPP PPP Eth Eth

PPP Header IP Header PPP Trailer

F Addr Ctrl Protocol S:10.0.0.1 D:10.0.1.9 CRC F

0xFF 0x0021 IP
0x8021 NCP for IP


IPover PPP and
IP over Ethernet
PPP
IP: 10.0.1.9
IP: 10.0.0.2 R IP: 10.0.1.1
Eth : B
Eth : A

IP IP
PPP PPP Eth Eth

PPP Header IP Header PPP Trailer

F Addr Ctrl Protocol S:10.0.0.1 D:10.0.1.9 CRC F

Ethernet header IP header Ethernet
0xFF 0x0021 IP
0x8021 NCP for IP P S:A D: B Type S:10.0.0.1 D:10.0.1.9 CRC

0x0800 IP

IP on LANs

 Problems to be solved
 How to encapsulate IP packets in frames ?
 How to find the LAN address of the IP destination ?
 LAN efficiently supports broadcast/multicast
transmission
 When a host needs to find the LAN address of
another IP host, it broadcasts a request
 The owner of the destination IP address will reply and
provided its LAN address
 LAN doesnʼt efficiently support broadcast/
multicast
 Maintain a server storing IP address:MAC address pairs
 Each host knows serverʼs MAC address and registers its
address pair
 Each host sends request to server to map IP addresses


Address Resolution Protocol

1
IP: 10.0.1.22 IP: 10.0.1.11 IP: 10.0.1.8 IP: 10.0.1.9
Eth : A Eth : E Eth : C Eth : B

10.0.1.22 needs to send an IP packet to 10.0.1.8



1
IP: 10.0.1.22 IP: 10.0.1.11 IP: 10.0.1.8 IP: 10.0.1.9


2
IP: 10.0.1.22 IP: 10.0.1.11 IP: 10.0.1.8 IP: 10.0.1.9

ARP : broadcast frame Addr Eth 10.0.1.8 ?



1
IP: 10.0.1.22 IP: 10.0.1.11 IP: 10.0.1.8 IP: 10.0.1.9


2
IP: 10.0.1.22 IP: 10.0.1.11 IP: 10.0.1.8 IP: 10.0.1.9

ARP : broadcast frame Addr Eth 10.0.1.8 ?

3
IP: 10.0.1.22 IP: 10.0.1.11 IP: 10.0.1.8 IP: 10.0.1.9

10.0.1.8 replies in an Ethernet frame and A knows the MAC address to send
its IP packet


ARP frame format

Preamble
[7 bytes]
Delimiter[1byte]

Destination
Address
Source
Address

Type: 0x806

Header

Sender MAC

Sender IP

Target MAC

Target IP

CNP3/2008.5. CRC [32 bits] © O. Bonaventure, 2008

ARP frame format

Preamble
[7 bytes]
Delimiter[1byte]

Destination
Broadcast : 111...111
Address
Source
Address

Type: 0x806

Header

Sender MAC

Sender IP

Target MAC

Target IP


ARP frame format

Preamble
[7 bytes]
Delimiter[1byte]

Destination
Address
Source MAC address of the sender
Address

Type: 0x806

Header

Sender MAC

Sender IP

Target MAC

Target IP


ARP frame format

Preamble
[7 bytes]
Delimiter[1byte]

Destination
Address
Source MAC address of the sender
Address
Common header for all ARP frames
Type: 0x806 - Hardware type Ethernet is 1
- Protocol type , IP is 0x0800.
Header
- Hardware length : length of MAC address
- Protocol length : length of network layer
address
Sender MAC - Operation : 1 for request, 2 for reply, 3 for
RARP request, and 4 for RARP reply.
Sender IP

Target MAC

Target IP


Optimisations

 When should a host send ARP requests ?
 Before sending each IP packet ?
 No, each host/router maintains an ARP table that contains
the mapping between IP addresses and Ethernet
addresses. An ARP request is only sent when the ARP
table is empty

 How to deal with hosts that change their
addresses ?
 Expiration timer is associated to each entry in the
ARP table
 Line of ARP table is removed upon timer expiration.
 Some implementations send an ARP request to revalidate
it before removing the line
 Some implementations remember when ARP lines have
been used to avoid removing an important entry

IP over Ethernet : Example

IP: 10.0.1.9/24
IP: 10.0.1.8/24 Default R: 10.0.1.1
IP: 10.0.1.22/24 Default R: 10.0.1.1
Default R: 10.0.1.1 Eth : B
Eth : C ARP Table
Eth : A ARP Table
ARP Table empty
empty
empty

IP: 10.0.1.1/24
Eth : R
ARP Table
empty
R

 Transmission of an IP packet from 10.0.1.22 to 10.0.1.9
 Transmission of an IP packet from 10.0.1.22 to 10.0.2.9


IPv6 over Ethernet
 Neighbour discovery / address resolution
1 IPv6: 1080:0:0:0:8:A IPv6: 1080:0:0:0:8:E Ipv6: 1080:0:0:0:8:C
Eth : A Eth : E Eth : C

1080:0:0:0:8:A wants to send a packet to 1080:0:0:0:8:C

2
IPv6: 1080:0:0:0:8:A IPv6: 1080:0:0:0:8:E Ipv6: 1080:0:0:0:8:C
Eth : A Eth : E Eth : C

Neighbor solicitation: Addr Eth 1080:0:0:0:8:C ? sent to IPv6 multicast address

3 IPv6: 1080:0:0:0:8:E
IPv6: 1080:0:0:0:8:A
Eth : A Eth : E Ipv6: 1080:0:0:0:8:C
Eth : C

Neighbor advertisement: 1080:0:0:0:8:C is reachable via Ethernet Add : C


ICMPv6 Neighbour Discovery
• Replacement for IPv4ʼs ARP
• Neighbour solicitation
• Sent to
Type : 135 Code:0 Checksum
Reserved

Target IPv6 Address

• Neighbour advertisement
RSO Reserved

Target IPv6 Address

Target link layer Address


• Sent to
The IPv6 address for which the link-layer Reserved
(e.g. Ethernet) address is needed. Target IPv6 Address
May also contain an optional ﬁeld with the link-layer
(e.g. Ethernet) address of the sender.

RSO Reserved

Target IPv6 Address



• Sent to

RSO Reserved

The IPv6 and link-layer addresses Target IPv6 Address



• Sent to

R : true if node is a router
S : true if answers to a neighbour solicitation Type : 136 Code:0 Checksum
RSO Reserved

The IPv6 and link-layer addresses Target IPv6 Address



IPv6 autoconfiguration

 How can a node obtain its IPv6 address ?
 Manual configuration
 From a server by using DHCPv6 as in IPv4
• Automatically
• Router advertises prefix on LAN by sending ICMPv6
messages to “all IPv6 hosts” multicast address
• Hosts build their address by concatenating the prefix
with their MAC Address converted in 64 bits format



• Automatically
R
Preﬁx= 2001:6a8:3080:1/64

Ethernet : 0800:200C:417A


• Automatically
R
Preﬁx= 2001:6a8:3080:1/64

IPv6 address : 2001:6a8:3080:1:M64(800:200C:417A)


Router advertisements

Ver Tclass Flow Label
Payload Length 58 255

Router IPv6 address
(link local)

FF02::1
(all nodes)

Type:134 Code : 0 Checksum
CurHLim M O Res Router lifetime
Reachable Time
Retrans Timer

Options


Maximum hop limit to avoid spoofed packets from
outside LAN
Ver Tclass Flow Label
Payload Length 58 255

Router IPv6 address
(link local)

FF02::1
(all nodes)

Reachable Time
Retrans Timer

Options


outside LAN
Ver Tclass Flow Label Value of hop limit to be used by hosts when
Payload Length 58 255 sending IPv6 packets

Router IPv6 address
(link local)

FF02::1
(all nodes)

Reachable Time
Retrans Timer

Options


outside LAN

Router IPv6 address The lifetime associated with the default router in
(link local) units of seconds. 0 is the router sending the
advertisement is not a default router.
FF02::1
(all nodes)

Reachable Time
Retrans Timer

Options


outside LAN

FF02::1
(all nodes)
The time, in milliseconds, that a node assumes a
Type:134 Code : 0 Checksum neighbour is reachable after having received a
CurHLim M O Res Router lifetime reachability conﬁrmation.
Reachable Time
Retrans Timer The time, in milliseconds, between retransmitted
Neighbor Solicitation messages.
Options


outside LAN

FF02::1
(all nodes)
The time, in milliseconds, that a node assumes a
Type:134 Code : 0 Checksum neighbour is reachable after having received a
CurHLim M O Res Router lifetime reachability conﬁrmation.
Reachable Time
Retrans Timer The time, in milliseconds, between retransmitted
Neighbor Solicitation messages.
Options
MTU to be used on the LAN
Preﬁxes to be used on the LAN


Router advertisements options
• Format of the options Type Length Options
Options (cont.)

• MTU option Type : 5 Length:1 Reserved
MTU

• Preﬁx option Type : 3 Length:4 PreLen L A Res.
Valid Lifetime
Preferred Lifetime
Reserved2
IPv6 preﬁx


Options (cont.)

MTU


Number of bits in IPv6 preﬁx that identify subnet Valid Lifetime
Preferred Lifetime
Reserved2
IPv6 preﬁx


Options (cont.)

MTU


Preferred Lifetime
Reserved2
The validity period of the preﬁx in seconds IPv6 preﬁx


Options (cont.)

MTU


Preferred Lifetime
Reserved2
The validity period of the prefix in seconds IPv6 prefix

The duration in seconds that addresses generated
from the prefix via stateless address
autoconfiguration remain preferred.


IPv6 autoconﬁguration (2)

 What happens when an endsystem boots ?
 It knows nothing about its current network
 but needs an IPv6 address to send ICMPv6 messages

R

FE80::M64(800:200C:417A) ))

 Use Link-local IPv6 address (FE80::/64)
 Each host, when it boots, has a link-local IPv6 address
 But another node might have chosen the same address !



 What happens when an endsystem boots ?
 It knows nothing about its current network
 but needs an IPv6 address to send ICMPv6 messages

R

ICMPv6 : Neighbour Solicitation
Sent to multicast address
Is someone using IPv6 address :
FE80::M64(800:200C:417A) ?
FE80::M64(800:200C:417A) ))
Address is valid if nobody answers
 Use Link-local IPv6 address (FE80::/64)
 Each host, when it boots, has a link-local IPv6 address
 But another node might have chosen the same address !


 How to obtain the IPv6 preﬁx of the subnet ?
• Wait for router advertisements (e.g. 30 seconds)
• Solicit router advertisement

R

ICMPv6 : Router Solicitation
IPv6 Src: FE80::M64(800:200C:417A)
IPv6 Dest: FF02::2
FE80::M64(800:200C:417A)



 Router will re-advertise preﬁx
R

FE80::M64(800:200C:417A)

 IPv6 addresses can be allocated for limited lifetime
 This allows IPv6 to easily support renumbering


 Router will re-advertise prefix
R

ICMPv6 : Router Advertisement
IPv6 Src: FE80::M64(EthernetR)
IPv6 Dest: FF02::1
IPv6 Prefix = 2001:6a8:1100::/48
Ethernet : 0800:200C:417A Prefix lifetime
FE80::M64(800:200C:417A)

 IPv6 addresses can be allocated for limited lifetime
 This allows IPv6 to easily support renumbering

Privacy issues with IPv6
address autoconfiguration
• Issue
• Autoconfigured IPv6 addresses contain the MAC
address of the hosts
• MAC addresses are fixed and unique
• A laptop/user could be identified by tracking the lower
64 bits of its IPv6 addresses

• How to maintain privacy with IPv6 ?
• Use DHCPv6 and configure server to never
reallocate the same IPv6 address
• Allow hosts to use random host ids in lower 64
bits of their IPv6 address
• algorithms have been implemented to generate such
random host ids on nodes with and without stable
storage


Ethernet today

 The coaxial cable is not used anymore

 Ethernet cabling today
 Structured twisted pair cabling
 Optical ﬁber for some point-to-point links

 Ethernet organisation
 Not anymore a bus
 Ethernet is now a star-shaped network !


Ethernet with structured cabling
 How to perform CSMA/CD in a star-shaped
network ?

Hub :
receives electrical signal on one port,
regenerates this signal and forwards it over all
other ports besides the port from which it
received it

Collision domain : set of stations that could be in collision

Hub and the reference model

 A hub is a relay operating the physical layer

Datalink Datalink
Physical Physical

Host A Hub Host B


Ethernet with structured cabling (2)

 How to build a larger Ethernet network ?
 Interconnect hubs together
Hub

Hub Hub

Hub

Issue : maximum 51.2 microsec
delay between any pair of stations

CNP3/2008.5. Collision domain : entire network © O. Bonaventure, 2008

Ethernet Switch
 Can we improve the performance of hubs ?
 Ethernet switch
 Operates in the datalink layer
 understands MAC address and ﬁlters frames based on
their addresses

Eth : C
Eth : A

Address Port
Src:A Dst:B
A West

B South

Eth : B C East

Ethernet Switch
 Can we improve the performance of hubs ?
 Ethernet switch
 Operates in the datalink layer
 understands MAC address and ﬁlters frames based on
their addresses

Eth : C
Eth : A

Address Port

A West

B South
Src:A Dst:B
Eth : B C East

Switch in the reference model

 A switch is a relay that operates in the
datalink layer

Network Network
Datalink Datalink
Physical Phys. Phys. Physical

Host A Switch Host B


Port-address table

 How to build the port-address table used by
Ethernet switches ?
 Manually
 Works in a lab, but Ethernet must be plug and
play
 Automatically
 Frame source address allows switch to learn the
location of hosts
 What happens when a destination address
cannot be found in the port-address table ?
 But be careful to age the information inside tables
as some hosts move from one port to another

 How to forward broadcast frames ?
 How to forward multicast frames ?

Frame processing

 Basic operation of an Ethernet switch
Arrival of frame F on port P
src=F.Source_Address;
dst=F.Destination_Address;
UpdateTable(src, P); // src heard on port P
if (dst==broadcast) || (dst is multicast)
{
for(Port p!=P) // forward all ports
ForwardFrame(F,p);
}
else
{
if(dst isin AddressPortTable)
{
ForwardFrame(F,AddressPortTable(dst));
}
else
{
for(Port p!=P) // forward all ports
ForwardFrame(F,p);
}
}

Network redundancy
 How to design networks that survive link and
node failures ?
 Add redundant switches
Address Port
A West
B South Eth : C

C East
Eth : A

Src:A Dst:C

Address Port
A West
Address Port B South
A North C North
B South Eth : B
CNP3/2008.5. C East © O. Bonaventure, 2008

Network redundancy
 How to design networks that survive link and
node failures ?
 Add redundant switches
Address Port
A West
B South Eth : C

C East Src:A Dst:C
Eth : A

Address Port
A West
Address Port B South
A North C North
B South Eth : B
CNP3/2008.5. C East © O. Bonaventure, 2008

Network redundancy (2)

 Does this always work ?
 Assume all switches have rebooted
Address Port

Eth : C

Eth : A

Src:A Dst:C

Address Port

Address Port

Eth : B

How to solve this problem ?



 The lawyerʼs way
 Add a sticker on all switches to indicate that they
must only be used in tree shaped networks and
should never ever be interconnected with loops



 The lawyerʼs way
 Add a sticker on all switches to indicate that they
must only be used in tree shaped networks and
should never ever be interconnected with loops

 The computer scientistʼs way
 Deﬁne a distributed algorithm that allows
switches to automatically discover the links
causing loops and remove them from the
topology


Principle of the solution

 Build a spanning tree inside network
 Each switch has a unique identiﬁer
 The switch with the lowest id is the root
 Disable all links that do not belong to spanning
tree
Switch 2

Switch 1
Switch 9

Switch 44

Switch 7 Switch 22

How to build the spanning tree

 Distributed algorithm run by switches
 Goals of the spanning tree protocol
 Elect the root of the spanning tree
 In practice, this will be the switch with the lowest id
 Compute the distance between each switch and
the root
 When several switches are attached to the same
LAN elect one forwarder and disable the others
 determine which ports/links should belong to the
spanning


Root and Designated Switches

 Root switch
 The Root Switch is the root of the spanning tree
 The Root switch may change upon the arrival of
new switches in the network
Switch 1
 Designated switch
 to avoid loops, only
one switch should
be responsible for Switch 9
Switch 4
forwarding frames
from the root on
any link
 Root switch is always
designated switch
for all its links

The switch identiﬁers



 Switch identiﬁers must be unique
 The easiest solution is to ask each manufacturer
to embed a unique Ethernet address on each
switch



switch

 But since the switch with the lowest identiﬁer
is the network root, network operators need
to inﬂuence the selection of the root switch



switch

 But since the switch with the lowest identifier
is the network root, network operators need
to influence the selection of the root switch
 64 bits switch identifier
 Upper 16 bits
 Priority defined by operator (default value : 32768)
 Lower 48 bits
 Unique Ethernet address assigned by manufacturer

The link costs

 Each switch port is attached to a link
 The costs of the links can be conﬁgured on
each link by the network operator
 Common guideline : Cost = 1000 / bandwidth

 Recommended values of link costs

Bandwidth Recommended Recommended
link cost range link cost value
10 Mbps 50-600 100
100 Mbps 10-60 19
1000 Mbps 3-10 4

Building the spanning tree

 802.1d protocol
 802.1d uses Bridge PDUs (BPDUs) containing
 Root ID : identiﬁer of the current root switch
 Cost : Cost of the shortest path between the switch
transmitting the BPDU and the root switch
 Transmitting ID : identiﬁer of the switch that transmits
the BPDU

 The BPDUs are sent by switches over their
attached LANs as multicast frames but they are
never forwarded
 switches that implement 802.1d listen to a special
Ethernet multicast group


Ordering of BPDUs

 BPDUs can be strictly ordered
 BPDU11[R=R1,C=C1, T=T1] is better than
BPDU2 [R=R2,C=C2, T=T2] if
 R1<R2
 R1=R2 and C1<C2
 R1=R2 and C1=C2 and T1<T2

 Example
BPDU1 BPDU2
R1 C1 T1 R2 C2 T2
29 15 35 31 12 32
35 80 39 35 80 40
35 15 80 35 18 38


Building the spanning tree (2)

 Behaviour of 802.1d protocol
 The root switch sends regularly BPDUs on all its
ports
 R=Root switch id, C=0, T= Root switch id
 Bootstrap
 If a switch does not receive BPDUs, it considers itself as root
and sends BPDUs
 On each port, a switch parses all the received
BPDUs and stores the best BPDU received on
each port
 Each switch can easily determiner the current root by
analysing all the BPDUs stored in its tables
 A switch stops sending BPDUs on a port if it
received a better BPDU on this port
 802.1d stabilises when a single switch sends a
BPDU over each LAN

802.1d port states
 802.1d port state based on received BPDUs
 Root port
 port on which the best 802.1d BPDU was received
 port used to receive the BPDUs sent by the root form
the shortest path
 A root port does not transmit BPDUs
 Only one root port on each switch
 Designated port
 port(s) used to send switchʼs BPDU upon reception of a
BPDU from the root via the Root port
 Switchʼs BPDU is
 current root, cost to reach root, switch identiﬁer
 0, one or more designated ports on each switch
 a port is designated if the switchʼs BPDU is better than
the best BPDU received on this port
 Blocked port (only receives 802.1d BPDUs)

802.1d port states (2)
 Example
 BPDUs received by switch 18
Root Cost Transmitter
port1 12 93 51
port2 12 85 47
port3 81 0 81
port4 15 31 27
 Root : switch 12
 port2 is the root port
 Switchʼs BPDU
 R=12, C=86, T=18
 This BPDU is better than the BPDUs received on
the other ports. They are thus designated


802.1d port states (3)

 Example
 BPDUs received by switch 92
Root Cost Transmitter
port1 81 0 81
port2 41 19 125
port3 41 12 315
port4 41 12 111
port5 41 13 90
 root : 41
 root port : port4
 Switchʼs BPDU
 R=41,C=13, T=92
 Port state
 port1 and port 2 : designated
 port 3 and port 5 : blocked

Port activity

 A port can be either active or inactive for
data frames
 Active port
 The switch captures Ethernet frames on its active ports
and forwards them over other ports (based on its own
port/address tables)
 The switch updates its port/address table based on the
frames received on this port
 Inactive port
 The switch does not listen to frames neither forward
frames on this port
 The port activity is ﬁxed once the spanning
tree has converged
 Root and designated ports become active
 Blocked ports become inactive
 Duration spanning tree computation, all ports are
CNP3/2008.5. inactive © O. Bonaventure, 2008

Port states and activity

Receive Transmit
BPDUs BPDUs
Blocked yes no
Root yes no
Designated yes yes

Learn Forward Data
Addresses Frames

Inactive no no
Active yes yes


Example network

 Compute the spanning tree in this network
by using 802.1d Switch 7

Switch 9
Switch 12

 Assume that 9 boots and then 12
CNP3/2008.5. and eventually 7 boots © O. Bonaventure, 2008

Impact of failures

 What kind of failures should be considered ?
 Failure (power-off) of the root switch
 A new root needs to be elected

 Failure of a designated switch
 Another switch should replace the designated one

 Failure of a link
 If the network is redundant, a disabled link should be
enabled to cope with the failure

 Failure of a link that disconnects the network
 We now have two different networks and a root switch
must be elected in each network


How to deal with failures ?

 Failure detection mechanisms
 Root switch sends its BPDU every Hello timer
and designated switches generate their own
BPDUs upon reception of this BPDU
 Default Hello timer is two seconds

 BPDUs stored in the switches age and are
removed when they timeout

 Failure notiﬁcation mechanism
 When a switch detects an important failure, it
sends a topology change (TC) BPDU to the Root
 Upon reception of a TC BPDU all switches stop
forwarding data frames and recompute spanning
CNP3/2008.5.
tree © O. Bonaventure, 2008

Ethernet Evolution

 Networks require higher bandwidth
 Fast Ethernet
 Physical layer
 bandwidth : 100 Mbps
 twisted pair or optical ﬁber
 No coaxial cable anymore

 MAC sublayer
 CSMA/CD unchanged
 minimum frame size : 512 bits
 slot time : 5.12 micro seconds
 Maximum distance : shorter than Ethernet 10 Mbps
 Same frame format as 10 Mbps Ethernet


Ethernet Evolution (2)

 Gigabit Ethernet
 Physical layer
 Bandwidth 1 Gbps
 Optical ﬁber or twisted pair

 MAC sublayer
 CSMA/CD still supported
 How was this achieved ?
 Two options
 Increase minimum frame size : not backward compatible with
Ethernet
 Reduce the maximum distance as for FastEthernet : but then
networks would have a diameter of 10 m
 Gigabit CSMA/CD hack
 minimum frame size is still 512 bits but the sender must
continue to send an electrical signal during the equivalent o
4096 bits
 same frame format as Ethernet
 but extensions allow to transmit Jumbo frames of up to 9KBytes

The Ethernet zoo

10BASE5 Thick coaxial cable, 500m

10BASE2 Thin coaxial cable, 185m

10BASE-T Two pairs of category 3+ UTP

10BASE-F 10 Mb/s over optical fiber

100BASE-TX Category 5 UTP or STP, 100 m maximum

100BASE-FX Two multimode optical fiber, 2 km maximum

1000BASE-CX Two pairs shielded twisted pair, 25m maximum

1000BASE-SX Two multimode or single mode optical fibers with lasers

10 Gbps optical fiber but also cat 6 twisted pair

40-100 Gbps being developed, standard expected in 2010, 40Gbps
one meter long for switch backplanes, 10 meters for copper
cable and 100 meters for fiber optics

Full duplex Ethernet

 Observations
 In many networks, Ethernet is a often a point-to-
point technology
 host-to-switch
 switch to switch

S1 S2

HUB
HUB

 Twisted-pairs and ﬁber-based physical layers
allow to send and receive at the same time

Ethernet full duplex (2)

 No collision is possible on a full duplex
Ethernet/FastEthernet/GigabitEthernet link
 Disable CSMA/CD on such links

 Advantages
 Improves bandwidth
 Both endpoints can transmit frames at the same time

 CSMA/CD is disabled
 No constraint on propagation delay anymore
 Ethernet network can be as large as we want !

 No constraint on minimum frame size anymore
 We do not need the frame extension hack for Gigabit Ethernet!


Full duplex Ethernet (3)

 Drawback
 If CSMA/CD is disabled, access control is
disabled and congestion can occur

Server
S1 S2
Client

FastEthernet (100 Mbps) Ethernet (10 Mbps)

 How to solve this problem inside Ethernet ?
 Add buffers to switches
 but inﬁnite buffers are impossible and useless anyway
 Cause collisions (e.g. jamming) to force collisions on the inter-
switch link and uplink is server is too fast
 Drawback : interswitch link could be entirely blocked
 Develop a new ﬂow control mechanism inside MAC layer
 Pause frame to slowdown transmission

Ethernet ﬂow control
server
Client

S1
FastEthernet Ethernet
(100 Mbps) (10 Mbps)

100 nsec
Frame2 [10000 bits] Frame1 [10000 bits]
Frame3 [10000 bits]
PAUSE [2msec] 1 microsec

Sender blocked

Frame2 [10000 bits]

 PAUSE frame indicates how much time the
upstream should wait before transmitting next frame

Ethernet ﬂow control
server
Client

S1
FastEthernet Ethernet
(100 Mbps) (10 Mbps)

100 nsec Frame1 [10000 bits]
Frame2 [10000 bits] Frame1 [10000 bits]
Frame3 [10000 bits]
PAUSE [2msec] 1 microsec

Sender blocked

Frame2 [10000 bits]

 PAUSE frame indicates how much time the
upstream should wait before transmitting next frame

Virtual LANs
 Allows to build several logical networks on
top of a single physical network
 Each port on each switch is
A B associated to a particular VLAN
 All the hosts that reside on the same
VLAN can exchange Ethernet
frames
 A host on VLAN1 cannot send an
C S F
Ethernet frame towards another
host that belongs to VLAN2
 Broadcast and multicast frames are
D E only sent to the members of the
VLAN

VLAN1 : A,E,F
VLAN2 : B,C,D

VLANs in campus networks
 How to support VLANs in a campus network

 Possible solutions
A
 Place on each switch a table
B
that maps each MAC address
on a VLAN id
 difficult to manage this table

C S1 S2 F  Change frame format used on
inter-switch links to include a
VLAN identifier
 new header added by first switch
D E  new header removed by last switch

VLAN1 : A,E,F
VLAN2 : B,C,D

VLAN frame format

 Used on inter-switch links
Destination
Address
Source
Address

Type

Payload

CRC [32 bits]

 Can also be used by trusted hosts (e.g. servers)
CNP3/2008.5. or routers © O. Bonaventure, 2008

VLAN frame format

Destination
Address
Source
Address
VLAN
Protocol Id
0x8100

Tag Control Info

Type

Payload

CRC [32 bits]


VLAN frame format

Destination
Address
Source
Address
VLAN Identiﬁes the frame as containing VLANtag
Protocol Id
0x8100

Tag Control Info

Type

Payload

CRC [32 bits]


VLAN frame format

Destination
Address
Source
Address
VLAN Identifies the frame as containing VLANtag
Protocol Id
0x8100 Tag control information contains two types
of information :
Tag Control Info
- VLAN identifier (12 bits) : up to 4094
different VLANs can be defined
Type - Priority (3 bits) : indicates the importance
of the frame and can be used by switches
to provide a better service for some frames
Payload (e.g. Voice)

CRC [32 bits]


Modes of operation

 Ad hoc
or Independent
Basic Service Set


Modes of operation

 Ad hoc
or Independent
Basic Service Set

 Infrastructure mode


Modes of operation

 Ad hoc
or Independent
Basic Service Set


BSS
BSS

The WiFi zoo

Standard Frequency Typical Raw Range in/out
throughput bandwidth (m)

802 .11 2.4 GHz 0.9 Mbps 2 Mbps 20 / 100

802 .11a 5 GHz 23 Mbps 54 Mbps 35 / 120

802 .11b 2.4 GHz 4.3 Mbps 11 Mbps 38 / 140

802 .11g 2.4 GHz 19 Mbps 54 Mbps 38 / 140

802 .11n 2.4 / 5 GHz 74 Mbps up to 600 70 / 250
Mbps


WiFi zoo and performance

 Performance issues with the multiple WiFi
transmission rates
 802.11, 802.11b and 802.11g operate on 2.4 GHz
frequency bands
 Many access points are multi-standard

54 Mbps



transmission rates
frequency bands

54 Mbps

Laptop supporting
only 802.11b


transmission rates
frequency bands

11 Mbps
54 Mbps

Laptop supporting
only 802.11b


transmission rates
frequency bands

11 Mbps
Laptop far away
54 Mbps

Laptop supporting
only 802.11b


transmission rates
frequency bands

1 Mbps

11 Mbps
Laptop far away
54 Mbps

Laptop supporting
only 802.11b

Practical issues
with WLAN deployments
 Home environment

 A WLAN can interfere with the neighbourʼs
CNP3/2008.5. WLAN © O. Bonaventure, 2008

Practical issues
with WLAN deployments
 Enterprise networks

 One access points can interfere with many other
CNP3/2008.5. access points © O. Bonaventure, 2008

The WiFi channel frequencies
 WiFi standards operate on several
frequencies called channels
 Usually about a dozen channels

 Why multiple channels ?
 Some channels my be affected by interference
and have a lower performance
 Some frequencies are reserved for speciﬁc
usage in some countries
 Allows frequency reuse when there are multiple
WiFi networks in the same area
 Unfortunately, many home access points operate by
default on the same factory set channel which causes
interference and reduced bandwidth


WLAN in enterprise
environments
 What could be done to improve the
performance of WLANs ?
 Reduce interference as much as possible
 Tune channel frequencies
 Reduce transmission power
 Similar to techniques used in GSM networks

 Recent deployments rely on centralised controllers
CNP3/2008.5. and thin access points © O. Bonaventure, 2008

Ring networks

S2
S1

unidirectional ring

S3
Interface

S5

S4
 Problem to be solved
 How to share fairly ring transmission capacity
among all devices attached to the ring ?

Ring networks (2)


Ring networks (2)

 How to share transmission capacity ?
 To avoid collisions, only one station should be
able to transmit a frame at any time
 The station that has the right to transmit must
own a special frame called token


Ring networks (2)

 How to share transmission capacity ?
 To avoid collisions, only one station should be
able to transmit a frame at any time
 The station that has the right to transmit must
own a special frame called token

 How can stations exchange token ?
 Token is a special frame that can be sent over
the ring network
 A station that needs to transmit a data frame can
 capture the token and remove it from the ring
 send one or more data frames
 send the token back on the ring to allow other stations
to capture it and transmit


Ring networks (3)

 Consequence
 When there are no data frames sent, stations
should continuously exchange the token
S1 S2
 How to achieve this ?
 A station must relay the
electrical signal it receives Unidirectional
ring
upstream when not S3
transmitting Interface
 it introducing a delay of one S5
bit transmission time
 If all stations behave so, S4
and token is small,
token will travel permanently
 If token is not small, increase the delay on the
token ring network

Ring networks (4)

 Data frame transmission
 A data frame requires a longer transmission time
than the ring delay

S1 S2

 Sender behaviour
 Captures token
 Sends data frame
 Removes data frame S3
from ring
 Sends token S5

S4


Ring networks in practice

 Two types of ring LANs
 Token Ring
 Invented by IBM
 Standardised by IEEE/ISO (802.5)
 Ring build with point-to-point twisted pair links
 4 Mbps
 16 Mbps
 Some work for 100 Mbps Token Ring
 Fiber Distributed Data Interface (FDDI)
 First data networks built with optical ﬁber
 standardised by ANSI
 100 Mbps
 up to 200 km and 1000 stations

 Other ring technologies exist and are used
 SONET/SDH
 DPT

Token Ring (1)

 Token
 travels permanently on ring when stations are
idle
 Size 24 bits
 Minimum delay on ring
 24 bits transmission times
 Actual ring delay
 Each station introduces a one-bit transmission time
delay
 Physical links have a propagation delay
 Each ring contains a monitor station that measures
delay during ring initialisation and adds delay if needed
 Interfaces
 Two modes of operation
 Listen : interface adds a one bit transmission delay
 Transmit : only if station owns the token

Token Ring (2)
 Frame format
 Token (24 bits)
 SD : starting delimiter
 invalid physical layer symbol with
Manchester coding 1 1 1
 AC : Access control SD AC ED
 ED : ending de ﬁn
 invalid physical layer symbol with
Manchester coding
 Data frame
1 1 1 (2or)6 (2 or)6 0... (nolimit) 4 1 1
SD AC FC Dest Source Data CRC ED FS
 FC : Frame control
 Allows to distinguish between control frames and data frames
 FS : Frame status

Token Ring (3)

Frame transmission
S2
Delay 1 bit S1
Listen mode

Unidirectional Ring

S3
Transmit mode
Interface

To From S5
Station Station

S4
 How to capture the token ?
 Rely on Token bit of AC ﬁeld and one bit delay

Token Ring (4)

 Whatʼs special about Token Ring
 Can efﬁciently support acknowledgements
 Frame Status contains two bits : A and C
 A and C are set to 0 when transmitting a frame
 When a receiver sees one frame destined to itself, it
sets A to 1
 When a receiver copies one frame destined to itself
inside its buffers, it sets C to 1

 Data frame (and FS) return to sender. By
checking A and C, it knows that :
 if A=0 and C=0, destination is down
 if A=1 and C=0, destination is up, but congested
 if A=1 and C=1, frame was received by destination


Token Ring (5)

 Issues with Token Ring
 How to ensure fairness ?
 A station should not be allowed to transmit indeﬁnitely
 Token Holding Time
 Maximum time during which a station can own the token and
transmit data frames without releasing the token
 Default : 10 milliseconds

 How to bootstrap the Token Ring ?
 Which station sends the ﬁrst token ?
 How to ensure that the Ring delay is long enough ?
 What happens when a station fails ?
 If it did not own the token, no issue
 If it owned the token while failing, then
 Which station will remove the current data frame from the
ring ?
 Which station will send the token on the ring ?

Token Ring (6)


Token Ring (6)

 How to bootstrap a Token Ring ?
 Complex problem
 Main idea
 One station should send the token
 The ﬁrst station on the ring hears nothing and notices
that there is a problem. It sends a special frame called
CLAIM_TOKEN
 If it receives the frame back, it becomes the monitor
 Each station must be able to become monitor


Token Ring (6)

 How to bootstrap a Token Ring ?
 Complex problem
 Main idea
 One station should send the token
 The ﬁrst station on the ring hears nothing and notices
that there is a problem. It sends a special frame called
CLAIM_TOKEN
 If it receives the frame back, it becomes the monitor
 Each station must be able to become monitor

 Monitorʼs responsibilities
 Ensure that token is never lost or corrupted
 Insert an artiﬁcial delay of 24 bit transmission
times on the ring
 Remove orphan and looping frames
 If the monitor fails, the ring must be bootstrapped
again © O. Bonaventure, 2008
CNP3/2008.5.

Token Ring (7)


Token Ring (7)

 Token surveillance
 Monitor checks how often its sees the token
 If there are N stations on the ring, then the monitor should
see the token at worst every N*THT seconds
 If token is lost, monitor cuts ring, removes electrical signal
and resend a new token


Token Ring (7)

 Orphan frames
 Frame with invalid coding or incomplete frame
 monitor cuts ring, removes electrical signal and resend a
new token


Token Ring (7)

 Orphan frames
 Frame with invalid coding or incomplete frame
 monitor cuts ring, removes electrical signal and resend a
new token
 Looping frames
 Every time monitor sees a frame, it sets its Monitor
bit of the AC ﬁeld to 1
 All stations send their frames with Monitor=0
 If a frame is seen twice by the monitor, it cuts ring,
removes electrical signal and resend a new token

FDDI

 Network topology FDDI
 Single ring like Token Ring
 Two counter rotating rings to deal with failures

S2
S1

S3

S5

Usually, only one ring is used
S4


FDDI


Failure

S2
S1

S3

S5

S4


FDDI


Bypass to reroute Failure
around failure

S2
S1

S3

S5

S4


FDDI (2)

 Token based access control
 A station can only transmit a data frame provided that it
owns the token
 Token Holding Time (THT)
 maximum duration of transmission
 Token Rotation Time (TRT)
 maximal delay for a token to rotate around the entire ring
 TRT ≤ Actives_Stations * THT + Ring_Latency

 When should the Token be released
 Immediately after removal of the data frame sent
 as in Token Ring
 Immediately after transmission of the data transfer,
without waiting for it to come back
 solution chosen for FDDI due to the high bandwidth and long
latency of the FDDI ring


FDDI (3)

 Delay sensitive service
 How to support two types of frames in FDDI ?
 normal data frames (asynchronous frames)
 example : ﬁle transfer, email, www
 delay sensitive data frames (synchronous frames)
 example : telephone, videoconference

 Solution
 Delay sensitive frames can be supported
provided that a FDDI ring can bound the
transmission delay of such a frame
 synchronous frames should be transmitted earlier than
normal frames on each station
 Since a station can always transmit when it captures
the token, a solution should bound the Token Rotation
Time to provide strict guarantees to delay sensitive
frames

FDDI (4)

 How to bound the TRT ?
 Target Token Rotation Time (TTRT)
 At ring initialisation, all stations propose their expected
TTRT and the smallest proposed value is chosen
 All stations must control their transmissions such that
the token rotation time is always smaller than TTRT
 each station measures the current TRT
 When a station captures the token, it can send its synchronous
frames
 there is a maximum amount of synchronous frames that can
be sent by each station. This maximum is negotiated by using
control frames.
 If after having sent synchronous frames TRT < TTRT, this
means that the token is circulating quickly and the station can
send asynchronous frames
 Otherwise the token must be released


FDDI (5)

 Frame format
Start of Frame
[8 bits]

Control [8 bits] Frame type (1 bit)
synchronous or asynchronous
48 bits addresses Destination Type of addresses
16 bits or 48 bits
Protocol Type information
Source 6 bits, similar to Type Ethernet DIX
special types for control frames and
token

Maximum 4478 bytes Data

CRC [32 bits]
End of Frame
[8 bits]
Similar to A and C bits in Token Ring
Status [24 bits]

Interconnection of Token Rings
 How to interconnect Token Ring networks ?

S1 S2 S7

S5
Bridge/Switch
S4 S9

 Possible solutions
 Use the spanning tree designed for Ethernet
CNP3/2008.5.
 Invent a new protocol © O. Bonaventure, 2008
 solution chosen by IBM for Token Ring

Interconnection of Token Rings (2)
S1
S7

S5

S9

 Basic idea
 Use source routing

 Problems
 How to identify the paths
 How to discover the paths ?

S1 S7

LAN1 LAN5
B3

S5
S9
B6
B1
LAN3
B4

 Identification of paths
 Each LAN has one unique identifier
 Each bridge has one identifier
 Each path is a list of pairs LAN#,bridge#



S1 S7

LAN1 B3
LAN5

S5

B6 S9

B1

B4 LAN3

 How to discover the path ?
 Control frame : all paths explorer
 Sent by source towards destination
 Forwarded by all bridges that add their identiﬁer and LAN
identiﬁer
 Destination sends back the ape frame to source by using
reverse path
 Each station caches the recent paths

Spanning Tree versus
Source Routing
 Spanning tree  Source routing
 complexity in  complexity in all
switches/bridges stations
 only a subset of the  the entire network is
network is used used
 entirely transparent  requires support on
stations
 multicast natively  spanning tree
supported required for multicast
 few control frames  many control frames
(802.1d) can be required


5 data link-lan.key

More Related Content

What's hot (20)

Viewers also liked (19)

Similar to 5 data link-lan.key (20)

More from Olivier Bonaventure (20)

Recently uploaded (20)

5 data link-lan.key

Editor's Notes