Presentation Slides
p  Will be available on
n  http://bgp4all.com/ftp/seminars/
APRICOT2016-BGP-Techniques.pdf
n  And on the APRICOT2016 website
p  Feel free to ask questions any time

BGP Techniques for Network Operators
p  BGP Basics
p  Scaling BGP
p  Using Communities
p  Deploying BGP in an ISP network

Border Gateway Protocol
p  A Routing Protocol used to exchange routing
information between different networks
n  Exterior gateway protocol
p  Described in RFC4271
n  RFC4276 gives an implementation report on BGP
n  RFC4277 describes operational experiences using BGP
p  The Autonomous System is the cornerstone of
BGP
n  It is used to uniquely identify networks with a common
routing policy

6
AS 100
Autonomous System (AS)
p  Collection of networks with same routing policy
p  Single routing protocol
p  Usually under single ownership, trust and
administrative control
p  Identified by a unique 32-bit integer (ASN)

Autonomous System Number
(ASN)
p  Two ranges
0-65535 (original 16-bit range)
65536-4294967295 (32-bit range – RFC6793)
p  Usage:
0 and 65535 (reserved)
1-64495 (public Internet)
64496-64511 (documentation – RFC5398)
64512-65534 (private use only)
23456 (represent 32-bit range in 16-bit world)
65536-65551 (documentation – RFC5398)
65552-4199999999 (public Internet)
4200000000-4294967295 (private use only – RFC6996)
p  32-bit range representation specified in RFC5396
n  Defines “asplain” (traditional format) as standard notation
7

Autonomous System Number
(ASN)
p  ASNs are distributed by the Regional Internet
Registries
n  They are also available from upstream ISPs who are
members of one of the RIRs
p  Current 16-bit ASN assignments up to 64297
have been made to the RIRs
n  Around 43000 16-bit ASNs are visible on the Internet
n  Around 200 left unassigned
p  Each RIR has also received a block of 32-bit ASNs
n  Out of 12400 assignments, around 9500 are visible on
the Internet
p  See www.iana.org/assignments/as-numbers
8

9
AS 100 AS 101
AS 102
E
B D
A C
Peering
BGP Basics
p  Runs over TCP – port 179
p  Path vector protocol
p  Incremental updates
p  “Internal” & “External” BGP

10
AS 100 AS 101
AS 102
DMZ
Network
A
B
C
D
E
p  DMZ is the link or network shared between ASes
Demarcation Zone (DMZ)

BGP General Operation
p  Learns multiple paths via internal and
external BGP speakers
p  Picks the best path and installs in the
forwarding table
p  Best path is sent to external BGP
neighbours
p  Policies are applied by influencing the best
path selection

eBGP & iBGP
p  BGP used internally (iBGP) and externally
(eBGP)
p  iBGP used to carry
n  Some/all Internet prefixes across ISP
backbone
n  ISP’s customer prefixes
p  eBGP used to
n  Exchange prefixes with other ASes
n  Implement routing policy

13
BGP/IGP model used in ISP
networks
p  Model representation
IGP
iBGP
eBGP eBGP eBGP
AS1 AS2 AS3 AS4
IGP
iBGP
IGP
iBGP
IGP
iBGP

14
AS 100 AS 101
C
A
B
External BGP Peering (eBGP)
p  Between BGP speakers in different AS
p  Should be directly connected
p  Never run an IGP between eBGP peers

Internal BGP (iBGP)
p  BGP peer within the same AS
p  Not required to be directly connected
n  IGP takes care of inter-BGP speaker
connectivity
p  iBGP speakers must to be fully meshed:
n  They originate connected networks
n  They pass on prefixes learned from outside the
ASN
n  They do not pass on prefixes learned from
other iBGP speakers

16
AS 100
A
D
C
B
Internal BGP Peering (iBGP)
p  Topology independent
p  Each iBGP speaker must peer with every other
iBGP speaker in the AS

17
Peering between Loopback Interfaces
p  Peer with loop-back interface
n  Loop-back interface does not go down – ever!
p  Do not want iBGP session to depend on state of
a single interface or the physical topology
AS 100
A
B
C

BGP Attributes
BGP’s policy tool kit

What Is an Attribute?
p  Part of a BGP Update
p  Describes the characteristics of prefix
p  Can either be transitive or non-transitive
p  Some are mandatory
19
Next
Hop AS Path ......MED...

BGP Attributes
p  Carry various information about or
characteristics of the prefix being
propagated
n  AS-PATH
n  NEXT-HOP
n  ORIGIN
n  AGGREGATOR
n  LOCAL_PREFERENCE
n  Multi-Exit Discriminator
n  (Weight)
n  COMMUNITY

AS-Path
p  Sequence of ASes a
route has traversed
p  Mandatory
transitive attribute
p  Used for:
n  Loop detection
n  Applying policy
21
AS 100
AS 300
AS 200
AS 500
AS 400
170.10.0.0/16 180.10.0.0/16
150.10.0.0/16
180.10.0.0/16 300 200 100
170.10.0.0/16 300 200
150.10.0.0/16 300 400
180.10.0.0/16 300 200 100
170.10.0.0/16 300 200

AS-Path (with 16 and 32-bit ASNs)
p  Internet with 16-bit
and 32-bit ASNs
n  32-bit ASNs are
65536 and above
p  AS-PATH length
maintained
22
180.10.0.0/16 300 23456 23456
170.10.0.0/16 300 23456
AS 70000
AS 300
AS 80000
AS 90000
AS 400
170.10.0.0/16 180.10.0.0/16
150.10.0.0/16
180.10.0.0/16 300 80000 70000
170.10.0.0/16 300 80000
150.10.0.0/16 300 400

AS-Path loop detection
p  180.10.0.0/16 is
not accepted by
AS100 as the prefix
has AS100 in its
AS-PATH – this is
loop detection in
action 23
AS 100
AS 300
AS 200
AS 500
170.10.0.0/16 180.10.0.0/16
180.10.0.0/16 300 200 100
170.10.0.0/16 300 200
140.10.0.0/16 300
140.10.0.0/16 500 300
170.10.0.0/16 500 300 200
140.10.0.0/16

Next Hop
p  eBGP – address of external
neighbour
p  iBGP – NEXT_HOP from eBGP
p  Mandatory non-transitive
attribute 24
160.10.0.0/16
150.10.0.0/16
150.10.1.1 150.10.1.2
AS 100
AS 300
AS 200
150.10.0.0/16 150.10.1.1
160.10.0.0/16 150.10.1.1
eBGP
iBGP
A B
C

iBGP Next Hop
p  Next hop is ibgp router loopback address
p  Recursive route look-up
25
AS 300
B
C
120.1.1.0/24 120.1.254.2
120.1.2.0/23 120.1.254.3
iBGP
120.1.1.0/24
A
D
Loopback
120.1.254.3/32
Loopback
120.1.254.2/32
120.1.2.0/23

Third Party Next Hop
p  eBGP between Router A
and Router B
p  eBGP between Router B
and Router C
p  120.68.1/24 prefix has
next hop address of
150.1.1.3 – this is used
by Router A instead of
150.1.1.2 as it is on
same subnet as Router B
p  More efficient
p  No extra config needed26
120.68.1.0/24
150.1.1.1
150.1.1.2
120.68.1.0/24 150.1.1.3
AS 200
AS 202
A
B C
AS 201
150.1.1.3

Next Hop Best Practice
p  BGP default is for external next-hop to be
propagated unchanged to iBGP peers
n  This means that IGP has to carry external
next-hops
n  Forgetting means external network is invisible
n  With many eBGP peers, it is unnecessary extra
load on IGP
p  ISP Best Practice is to change external
next-hop to be that of the local router

Next Hop (Summary)
p  IGP should carry route to next hops
p  Recursive route look-up
p  Unlinks BGP from actual physical topology
p  Change external next hops to that of local
router
p  Allows IGP to make intelligent forwarding
decision

Origin
p  Conveys the origin of the prefix
p  Historical attribute
n  Used in transition from EGP to BGP
p  Transitive and Mandatory Attribute
p  Influences best path selection
p  Three values: IGP, EGP, incomplete
n  IGP – generated by BGP network statement
n  EGP – generated by EGP
n  incomplete – redistributed from another
routing protocol

Aggregator
p  Conveys the IP address of the router or
BGP speaker generating the aggregate
route
p  Optional & transitive attribute
p  Useful for debugging purposes
p  Does not influence best path selection

Local Preference
31
AS 400
AS 200
160.10.0.0/16
AS 100
AS 300
E
B
C
A
D

Local Preference
32
AS 400
AS 200
160.10.0.0/16
AS 100
AS 300
500 800 E
B
C
A
D

Local Preference
33
AS 400
AS 200
160.10.0.0/16
AS 100
AS 300
500 800 E
B
C
A
D
160.10.0.0/16 500
> 160.10.0.0/16 800

Local Preference
34
AS 400
AS 200
160.10.0.0/16
AS 100
AS 300
500 800 E
B
C
A
D
160.10.0.0/16 500
> 160.10.0.0/16 800

Local Preference
p  Non-transitive and optional attribute
p  Local to an AS – non-transitive
n  Default local preference is 100 (Cisco IOS)
p  Used to influence BGP path selection
n  determines best path for outbound traffic
p  Path with highest local preference wins

Multi-Exit Discriminator (MED)
36
AS 400
AS 200
120.68.1.0/24
A B
C D

37
AS 400
AS 200
120.68.1.0/24
A B
120.68.1.0/24 1000120.68.1.0/24 2000
C D

38
AS 400
AS 200
120.68.1.0/24
A B
120.68.1.0/24 1000120.68.1.0/24 2000
C D
120.68.1.0/24 2000
> 120.68.1.0/24 1000

39
AS 400
AS 200
120.68.1.0/24
A B
120.68.1.0/24 1000120.68.1.0/24 2000
C D
120.68.1.0/24 2000
> 120.68.1.0/24 1000

Multi-Exit Discriminator
p  Inter-AS – non-transitive & optional attribute
p  Used to convey the relative preference of entry
points
n  Determines best path for inbound traffic
p  Comparable if paths are from same AS
n  Implementations have a knob to allow comparisons of
MEDs from different ASes
p  Path with lowest MED wins
p  Absence of MED attribute implies MED value of
zero (RFC4271)

Multi-Exit Discriminator
“metric confusion”
p  MED is non-transitive and optional attribute
n  Some implementations send learned MEDs to iBGP peers
by default, others do not
n  Some implementations send MEDs to eBGP peers by
default, others do not
p  Default metric varies according to vendor
implementation
n  Original BGP spec (RFC1771) made no recommendation
n  Some implementations handled absence of metric as
meaning a metric of 0
n  Other implementations handled the absence of metric as
meaning a metric of 232-1 (highest possible) or 232-2
n  Potential for “metric confusion”

Community
p  Communities are described in RFC1997
n  Transitive and Optional Attribute
p  32 bit integer
n  Represented as two 16 bit integers (RFC1998)
n  Common format is <local-ASN>:xx
n  0:0 to 0:65535 and 65535:0 to 65535:65535 are
reserved
p  Used to group destinations
n  Each destination could be member of multiple
communities
p  Very useful in applying policies within and
between ASes

Community Example
(before)
43
permit 160.10.0.0/16 out
ISP 1
AS 300
Upstream
AS 400F
E
D
permit 160.10.0.0/16 in
AS 100
A
160.10.0.0/16
C

Community Example
(before)
44
permit 160.10.0.0/16 out
ISP 1
AS 300
Upstream
AS 400F
E
AS 200
permit 170.10.0.0/16 in
B
170.10.0.0/16
D
permit 160.10.0.0/16 in
AS 100
A
160.10.0.0/16
C

Community Example
(before)
45
permit 160.10.0.0/16 out
ISP 1
AS 300
Upstream
AS 400F
E
permit 170.10.0.0/16 out
AS 200
permit 170.10.0.0/16 in
B
170.10.0.0/16
D
permit 160.10.0.0/16 in
AS 100
A
160.10.0.0/16
C

Community Example
(before)
46
permit 160.10.0.0/16 out
ISP 1
permit 100.10.0.0/16 in
X
Peer AS1
100.10.0.0/16
AS 300
Upstream
AS 400F
E
permit 170.10.0.0/16 out
AS 200
permit 170.10.0.0/16 in
B
170.10.0.0/16
D
permit 160.10.0.0/16 in
AS 100
A
160.10.0.0/16
C

Community Example
(after)
47
160.10.0.0/16 300:1
ISP 1
AS 300
Upstream
AS 400F
E
160.10.0.0/16 300:1
AS 100
A
160.10.0.0/16
C
D

Community Example
(after)
48
160.10.0.0/16 300:1
ISP 1
AS 300
Upstream
AS 400F
E
AS 200
170.10.0.0/16 300:1
B
170.10.0.0/16
160.10.0.0/16 300:1
AS 100
A
160.10.0.0/16
C
D

Community Example
(after)
49
160.10.0.0/16 300:1
ISP 1
AS 300
Upstream
AS 400F
E
170.10.0.0/16 300:1
AS 200
170.10.0.0/16 300:1
B
170.10.0.0/16
160.10.0.0/16 300:1
AS 100
A
160.10.0.0/16
C
D

Community Example
(after)
50
160.10.0.0/16 300:1
ISP 1
100.10.0.0/16 300:9
X
Peer AS1
100.10.0.0/16
AS 300
Upstream
AS 400F
E
170.10.0.0/16 300:1
AS 200
170.10.0.0/16 300:1
B
170.10.0.0/16
160.10.0.0/16 300:1
AS 100
A
160.10.0.0/16
C
D

Well-Known Communities
p  Several well known communities
n  www.iana.org/assignments/bgp-well-known-communities
p  no-export 65535:65281
n  do not advertise to any eBGP peers
p  no-advertise 65535:65282
n  do not advertise to any BGP peer
p  no-export-subconfed 65535:65283
n  do not advertise outside local AS (only used with
confederations)
p  no-peer 65535:65284
n  do not advertise to bi-lateral peers (RFC3765)
51

No-Export Community
p  AS100 announces aggregate and subprefixes
n  Intention is to improve loadsharing by leaking subprefixes
p  Subprefixes marked with no-export community
p  Router G in AS200 does not announce prefixes with no-
export community set
52
105.7.0.0/16
105.7.X.X no-export
105.7.0.0/16
AS 100 AS 200
105.7.X.X
C F
G
DA
B E

No-Peer Community
53
p  Sub-prefixes marked with no-peer community are
not sent to bi-lateral peers
n  They are only sent to upstream providers
105.7.0.0/16
105.7.X.X no-peer
A
B
E
D
C
C&D&E
are peers
e.g.
Tier-1s
upstream
upstream
upstream
105.7.0.0/16
105.7.0.0/16

What about 4-byte ASNs?
p  Communities are widely used for encoding ISP
routing policy
n  32 bit attribute
p  RFC1998 format is now “standard” practice
n  ASN:number
p  Fine for 2-byte ASNs, but 4-byte ASNs cannot be
encoded
p  Solutions:
n  Use “private ASN” for the first 16 bits
n  Wait for http://datatracker.ietf.org/doc/draft-ietf-idr-
as4octet-extcomm-generic-subtype/ to be implemented
54

Community
Implementation details
p  Community is an optional attribute
n  Some implementations send communities to
iBGP peers by default, some do not
n  Some implementations send communities to
eBGP peers by default, some do not
p  Being careless can lead to community
“confusion”
n  ISPs need consistent community policy within
their own networks
n  And they need to inform peers, upstreams and
customers about their community expectations

BGP Path Selection
Algorithm
Why Is This the Best Path?

BGP Path Selection Algorithm for
Cisco IOS: Part One
1.  Do not consider path if no route to next
hop
2.  Do not consider iBGP path if not
synchronised (Cisco IOS)
3.  Highest weight (local to router)
4.  Highest local preference (global within
AS)
5.  Prefer locally originated route
6.  Shortest AS path
57

Cisco IOS: Part Two
7.  Lowest origin code
n  IGP < EGP < incomplete
8.  Lowest Multi-Exit Discriminator (MED)
n  If bgp deterministic-med, order the paths by
AS number before comparing
n  If bgp always-compare-med, then compare
for all paths
n  Otherwise MED only considered if paths are
from the same AS (default)
58

Cisco IOS: Part Three
9.  Prefer eBGP path over iBGP path
10.  Path with lowest IGP metric to next-hop
11.  For eBGP paths:
n  If multipath is enabled, install N parallel paths
in forwarding table
n  If router-id is the same, go to next step
n  If router-id is not the same, select the oldest
path
59

Cisco IOS: Part Four
12.  Lowest router-id (originator-id for
reflected routes)
13.  Shortest cluster-list
n  Client must be aware of Route Reflector
attributes!
14.  Lowest neighbour address
60

BGP Path Selection Algorithm
p  In multi-vendor environments:
n  Make sure the path selection processes are
understood for each brand of equipment
n  All have to follow the RFC, but because of
“customer demand”, each vendor has:
p  Slightly different implementations
p  Extra steps
p  Extra features
n  Watch out for possible MED confusion

Applying Policy with BGP
Controlling Traffic Flow & Traffic
Engineering

Applying Policy in BGP:
Why?
p  Network operators rarely “plug in routers
and go”
p  External relationships:
n  Control who they peer with
n  Control who they give transit to
n  Control who they get transit from
p  Traffic flow control:
n  Efficiently use the scarce infrastructure
resources (external link load balancing)
n  Congestion avoidance
n  Terminology: Traffic Engineering

Applying Policy in BGP:
How?
p  Policies are applied by:
n  Setting BGP attributes (local-pref, MED, AS-
PATH, community), thereby influencing the
path selection process
n  Advertising or Filtering prefixes
n  Advertising or Filtering prefixes according to
ASN and AS-PATHs
n  Advertising or Filtering prefixes according to
Community membership

Applying Policy with BGP:
Tools
p  Most implementations have tools to apply
policies to BGP:
n  Prefix manipulation/filtering
n  AS-PATH manipulation/filtering
n  Community Attribute setting and matching
p  Implementations also have policy
language which can do various match/set
constructs on the attributes of chosen BGP
routes

BGP Capabilities
Extending BGP

BGP Capabilities
p  Documented in RFC2842
p  Capabilities parameters passed in BGP
open message
p  Unknown or unsupported capabilities will
result in NOTIFICATION message
p  Codes:
n  0 to 63 are assigned by IANA by IETF
consensus
n  64 to 127 are assigned by IANA “first come
first served”
n  128 to 255 are vendor specific

BGP Capabilities
p  Current capabilities are:
See www.iana.org/assignments/capability-codes
0 Reserved [RFC3392]
1 Multiprotocol Extensions for BGP-4 [RFC4760]
2 Route Refresh Capability for BGP-4 [RFC2918]
3 Outbound Route Filtering Capability [RFC5291]
4 Multiple routes to a destination capability [RFC3107]
5 Extended Next Hop Encoding [RFC5549]
64 Graceful Restart Capability [RFC4724]
65 Support for 4 octet ASNs [RFC6793]
66 Deprecated
67 Support for Dynamic Capability [ID]
68 Multisession BGP [ID]
69 Add Path Capability [ID]
70 Enhanced Route Refresh Capability [RFC7313]
71 Long Lived Graceful Restart [ID]
72 CP-ORF Capability [RFC7543]
73 FQDN Capability [ID]

BGP Capabilities
p  Multiprotocol extensions
n  This is a whole different world, allowing BGP to support
more than IPv4 unicast routes
n  Examples include: v4 multicast, IPv6, v6 multicast,
VPNs
n  Another tutorial (or many!)
p  Route refresh is a well known scaling technique –
covered shortly
p  32-bit ASNs arrived in 2006
p  The other capabilities are still in development or
not widely implemented or deployed yet

BGP for Internet Service Providers
p  BGP Basics
p  Scaling BGP
p  Using Communities
p  Deploying BGP in an ISP network

BGP Scaling Techniques
p  Original BGP specification and
implementation was fine for the Internet
of the early 1990s
n  But didn’t scale
p  Issues as the Internet grew included:
n  Scaling the iBGP mesh beyond a few peers?
n  Implement new policy without causing flaps
and route churning?
n  Keep the network stable, scalable, as well as
simple?

BGP Scaling Techniques
p  Current Best Practice Scaling Techniques
n  Route Refresh
n  Route Reflectors (and Confederations)
p  Deploying 4-byte ASNs
p  Deprecated Scaling Techniques
n  Route Flap Damping

Dynamic Reconfiguration
Route Refresh

Route Refresh
p  BGP peer reset required after every policy
change
n  Because the router does not store prefixes
which are rejected by policy
p  Hard BGP peer reset:
n  Tears down BGP peering & consumes CPU
n  Severely disrupts connectivity for all networks
p  Soft BGP peer reset (or Route Refresh):
n  BGP peering remains active
n  Impacts only those prefixes affected by policy
change 75

Route Refresh Capability
p  Facilitates non-disruptive policy changes
p  For most implementations, no
configuration is needed
n  Automatically negotiated at peer establishment
p  No additional memory is used
p  Requires peering routers to support “route
refresh capability” – RFC2918
n  Today most vendors do, and some do an
automatic route-refresh after BGP Policy
changes

Dynamic Reconfiguration
p  Use Route Refresh capability
n  Supported on virtually all routers
n  Find out from “show ip bgp neighbor”
n  Non-disruptive, “Good For the Internet”
p  Only hard-reset a BGP peering as a last
resort
77
Consider the impact to be
equivalent to a router reboot

Route Reflectors
Scaling the iBGP mesh

Scaling iBGP mesh
p  Avoid ½n(n-1) iBGP mesh
79
n=1000 ⇒ nearly
half a million
ibgp sessions!
14 routers = 91
iBGP sessions
p  Two solutions
n  Route reflector – simpler to deploy and run
n  Confederation – more complex, has corner case
advantages

Route Reflector: Principle
80
AS 100
A
CB

Route Reflector: Principle
81
AS 100
A
CB
Route Reflector

Route Reflector
p  Reflector receives
path from clients and
non-clients
p  Selects best path
p  If best path is from
client, reflect to other
clients and non-
clients
p  If best path is from
non-client, reflect to
clients only
p  Non-meshed clients
82
AS 100
A
B C
Clients
Reflectors

Route Reflector: Topology
p  Divide the backbone into multiple clusters
p  At least one route reflector and few clients
per cluster
p  Route reflectors are fully meshed
p  Clients in a cluster could be fully meshed
p  Single IGP to carry next hop and local
routes

Route Reflector: Loop Avoidance
p  Originator_ID attribute
n  Carries the RID of the originator of the route in
the local AS (created by the RR)
p  Cluster_list attribute
n  The local cluster-id is added when the update
is sent by the RR
n  Best to set cluster-id from router-id (address
of loopback)
n  (Some ISPs use their own cluster-id
assignment strategy – but needs to be well
documented!)

Route Reflector: Redundancy
p  Multiple RRs can be configured in the
same cluster – not advised!
n  All RRs in the cluster must have the same
cluster-id (otherwise it is a different cluster)
p  A router may be a client of RRs in different
clusters
n  Common today in ISP networks to overlay two
clusters – redundancy achieved that way
n  → Each client has two RRs = redundancy

Route Reflectors:
Redundancy
86
AS 100
Cluster One
Cluster Two
PoP2PoP1
PoP3

Route Reflector: Benefits
p  Solves iBGP mesh problem
p  Packet forwarding is not affected
p  Normal BGP speakers co-exist
p  Multiple reflectors for redundancy
p  Easy migration
p  Multiple levels of route reflectors

Route Reflector: Deployment
p  Where to place the route reflectors?
n  Always follow the physical topology!
n  This will guarantee that the packet forwarding
won’t be affected
p  Typical Service Provider network:
n  PoP has two core routers
n  Core routers are RR for the PoP
n  Two overlaid clusters

Route Reflector: Migration
p  Typical ISP network:
n  Core routers have fully meshed iBGP
n  Create further hierarchy if core mesh too big
p  Split backbone into regions
p  Configure one cluster pair at a time
n  Eliminate redundant iBGP sessions
n  Place maximum one RR per cluster
n  Easy migration, multiple levels

Route Reflectors: Migration
p  Migrate small parts of the network, one part at
a time. 92
AS 200
AS 100
AS 300
A
B
G
F
E
D
C

Confederations
p  Divide the AS into sub-AS
n  eBGP between sub-AS, but some iBGP
information is kept
p  Preserve NEXT_HOP across the
sub-AS (IGP carries this information)
p  Preserve LOCAL_PREF and MED
p  Usually a single IGP

Confederations (Cont.)
p  Visible to outside world as single AS –
“Confederation Identifier”
n  Each sub-AS uses a number from the private
AS range (64512-65534)
p  iBGP speakers in each sub-AS are fully
meshed
n  The total number of neighbours is reduced by
limiting the full mesh requirement to only the
peers in the sub-AS
n  Can also use Route-Reflector within sub-AS

Confederations
96
AS 200
Sub-AS
65530
Sub-AS
65532 Sub-AS
65531
C B
A
p  Configuration (Router C):
router bgp 65532
bgp confederation identifier 200
bgp confederation peers 65530 65531
neighbor 141.153.12.1 remote-as 65530
neighbor 141.153.17.2 remote-as 65531

Confederations: AS-Sequence
Sub-AS
65002
Sub-AS
65003
Sub-AS
65001
Confederation
100
G
Sub-AS
65004
C
D E
B
180.10.0.0/16 200
180.10.0.0/16 {65002} 200
A
180.10.0.0/16 {65004 65002} 200
H F
180.10.0.0/16 100 200

Route Propagation Decisions
p  Same as with “normal” BGP:
n  From peer in same sub-AS → only to external
peers
n  From external peers → to all neighbors
p  “External peers” refers to
n  Peers outside the confederation
n  Peers in a different sub-AS
n  Preserve LOCAL_PREF, MED and NEXT_HOP

Internet
Connectivity
Multi-Level
Hierarchy
Policy
Control Scalability
Migration
Complexity
Confederations
Route
Reflectors
Anywher
e in the
Network
Yes Yes
Yes
RRs or Confederations
Yes
Anywhere
in the
Network
Medium
Very High Very Low
Medium
to High
Most new service provider networks now deploy
Route Reflectors from Day One

More points about Confederations
p  Can ease “absorbing” other ISPs into you
ISP – e.g., if one ISP buys another
n  Or can use AS masquerading feature available
in some implementations to do a similar thing
p  Can use route-reflectors with
confederation sub-AS to reduce the sub-
AS iBGP mesh

Deploying 32-bit ASNs
How to support customers using
the extended ASN range

32-bit ASNs
p  Standards documents
n  Description of 32-bit ASNs
p  www.rfc-editor.org/rfc/rfc6793.txt
n  Textual representation
n  New extended community
p  AS 23456 is reserved as interface between
16-bit and 32-bit ASN world

32-bit ASNs – terminology
p  16-bit ASNs
n  Refers to the range 0 to 65535
p  32-bit ASNs
n  (or the extended range)
p  32-bit ASN pool

Getting a 32-bit ASN
p  Nowadays:
n  Standard application process to the RIRs
n  Or via upstream provider
n  Sample RIR policy
p  www.apnic.net/docs/policy/asn-policy.html
p  Bootstrap phase from 2007-2010
n  From 1st January 2007
p  32-bit ASNs were available on request
p  32-bit ASNs were assigned by default
p  16-bit ASNs were only available on request
p  No distinction – ASNs assigned from the 32-bit pool
104

Representation (1)
p  Initially three formats proposed for the
0-4294967295 ASN range :
n  asplain
n  asdot
n  asdot+
p  In reality:
n  Most operators favour traditional plain format
n  A few prefer dot notation (X.Y):
p  asdot for 65536-4294967295, e.g 2.4
p  asdot+ for 0-4294967295, e.g 0.64513
n  But regular expressions will have to be completely
rewritten for asdot and asdot+ !!!
105

Representation (2)
p  Rewriting regular expressions for asdot/asdot+
notation
p  Example:
n  ^[0-9]+$ matches any ASN (16-bit and asplain)
n  This and equivalents extensively used in BGP
multihoming configurations for traffic engineering
p  Equivalent regexp for asdot is:
n  ^([0-9]+)|([0-9]+.[0-9]+)$
p  Equivalent regexp for asdot+ is:
n  ^[0-9]+.[0-9]+$
106

Changes
p  32-bit ASNs are backward compatible with 16-bit ASNs
p  There is no flag day
p  You do NOT need to:
n  Throw out your old routers
n  Replace your 16-bit ASN with a 32-bit ASN
p  You do need to be aware that:
n  Your customers will come with 32-bit ASNs
n  ASN 23456 is not a bogon!
n  You will need a router supporting 32-bit ASNs to use a 32-bit
ASN locally
p  If you have a proper BGP implementation, 32-bit ASNs will
be transported silently across your network

How does it work?
p  If local router and remote router supports
configuration of 32-bit ASNs
n  BGP peering is configured as normal using the
32-bit ASN
p  If local router and remote router does not
support configuration of 32-bit ASNs
n  BGP peering can only use a 16-bit ASN
p  If local router only supports 16-bit ASN
and remote router/network has a 32-bit
ASN
n  Compatibility mode is initiated…

Compatibility Mode (1)
p  Local router only supports 16-bit ASN and
remote router uses 32-bit ASN
p  BGP peering initiated:
n  Remote asks local if 32-bit supported (BGP
capability negotiation)
n  When local says “no”, remote then presents
AS23456
n  Local needs to be configured to peer with
remote using AS23456
p  ⇒ Operator of local router has to configure
BGP peering with AS23456 109

Compatibility Mode (2)
p  BGP peering initiated (cont):
n  BGP session established using AS23456
n  32-bit ASN included in a new BGP attribute
called AS4_PATH
p  (as opposed to AS_PATH for 16-bit ASNs)
p  Result:
n  16-bit ASN world sees 16-bit ASNs and 23456
standing in for each 32-bit ASN
n  32-bit ASN world sees 16 and 32-bit ASNs
110

Example:
p  Internet with 32-
bit and 16-bit
ASNs
p  AS-PATH length
maintained
111
180.10.0.0/16 123 23456 23456
170.10.0.0/16 123 23456
AS 80000
AS 123
AS 70000
AS 90000
AS 321
170.10.0.0/16 180.10.0.0/16
150.10.0.0/16
180.10.0.0/16 123 70000 80000
170.10.0.0/16 123 70000
150.10.0.0/16 123 321

What has changed?
p  Two new BGP attributes:
n  AS4_PATH
p  Carries 32-bit ASN path info
n  AS4_AGGREGATOR
p  Carries 32-bit ASN aggregator info
n  Well-behaved BGP implementations will simply
pass these along if they don’t understand them
p  AS23456 (AS_TRANS)

asdot
format
asplain
format
What do they look like?
p  IPv4 prefix originated by AS196613
as4-7200#sh ip bgp 145.125.0.0/20
BGP routing table entry for 145.125.0.0/20, version
58734
Paths: (1 available, best #1, table default)
131072 12654 196613
204.69.200.25 from 204.69.200.25 (204.69.200.25)
Origin IGP, localpref 100, valid, internal, best
p  IPv4 prefix originated by AS3.5
as4-7200#sh ip bgp 145.125.0.0/20
58734
Paths: (1 available, best #1, table default)
2.0 12654 3.5
204.69.200.25 from 204.69.200.25 (204.69.200.25)
Origin IGP, localpref 100, valid, internal, best

Transition
AS
What do they look like?
p  IPv4 prefix originated by AS196613
n  But 16-bit AS world view:
BGP-view1>sh ip bgp 145.125.0.0/20
113382
Paths: (1 available, best #1, table Default-IP-Routing-
Table)
23456 12654 23456
204.69.200.25 from 204.69.200.25 (204.69.200.25)
Origin IGP, localpref 100, valid, external, best

If 32-bit ASN not supported:
p  Inability to distinguish between peer ASes using
32-bit ASNs
n  They will all be represented by AS23456
n  Could be problematic for transit provider’s policy
n  Workaround: use BGP communities instead
p  Inability to distinguish prefix’s origin AS
n  How to tell whether origin is real or fake?
n  The real and fake both represented by AS23456
n  (There should be a better solution here!)
115

If 32-bit ASN not supported:
p  Incorrect NetFlow summaries:
n  Prefixes from 32-bit ASNs will all be summarised under
AS23456
n  Traffic statistics need to be measured per prefix and
aggregated
n  Makes it hard to determine peerability of a neighbouring
network
p  Unintended filtering by peers and upstreams:
n  Even if IRR supports 32-bit ASNs, not all tools in use can
support
n  ISP may not support 32-bit ASNs which are in the IRR –
and don’t realise that AS23456 is the transition AS
116

Implementations (May 2011)
p  Cisco IOS-XR 3.4 onwards
p  Cisco IOS-XE 2.3 onwards
p  Cisco IOS 12.0(32)S12, 12.4(24)T, 12.2SRE, 12.2(33)SXI1
onwards
p  Cisco NX-OS 4.0(1) onwards
p  Quagga 0.99.10 (patches for 0.99.6)
p  OpenBGPd 4.2 (patches for 3.9 & 4.0)
p  Juniper JunOSe 4.1.0 & JunOS 9.1 onwards
p  Redback SEOS
p  Force10 FTOS7.7.1 onwards
p  http://as4.cluepon.net/index.php/Software_Support used
to have a complete list
117

Route Flap Damping
Network Stability for the 1990s
Network Instability for the 21st
Century!

Route Flap Damping
p  For many years, Route Flap Damping was
a strongly recommended practice
p  Now it is strongly discouraged as it
appears to cause far greater network
instability than it cures
p  But first, the theory…

Route Flap Damping
p  Route flap
n  Going up and down of path or change in
attribute
p  BGP WITHDRAW followed by UPDATE = 1 flap
p  eBGP neighbour going down/up is NOT a flap
n  Ripples through the entire Internet
n  Wastes CPU
p  Damping aims to reduce scope of route
flap propagation

Route Flap Damping (continued)
p  Requirements
n  Fast convergence for normal route changes
n  History predicts future behaviour
n  Suppress oscillating routes
n  Advertise stable routes
p  Implementation described in RFC 2439

Operation
p  Add penalty (1000) for each flap
n  Change in attribute gets penalty of 500
p  Exponentially decay penalty
n  Half life determines decay rate
p  Penalty above suppress-limit
n  Do not advertise route to BGP peers
p  Penalty decayed below reuse-limit
n  Re-advertise route to BGP peers
n  Penalty reset to zero when it is half of reuse-
limit

Operation
Reuse limit
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
0
1000
2000
3000
4000
Time
Penalty
Suppress limit
Network
Announced
Network
Re-announced
Network
Not Announced
Penalty

Operation
p  Only applied to inbound announcements
from eBGP peers
p  Alternate paths still usable
p  Controllable by at least:
n  Half-life
n  reuse-limit
n  suppress-limit
n  maximum suppress time

Configuration
p  Implementations allow various policy
control with flap damping
n  Fixed damping, same rate applied to all
prefixes
n  Variable damping, different rates applied to
different ranges of prefixes and prefix lengths

Route Flap Damping History
p  First implementations on the Internet by
1995
p  Vendor defaults too severe
n  RIPE Routing Working Group recommendations
in ripe-178, ripe-210, and ripe-229
n  http://www.ripe.net/ripe/docs
n  But many ISPs simply switched on the vendors’
default values without thinking

Serious Problems:
p  "Route Flap Damping Exacerbates Internet
Routing Convergence“
n  Zhuoqing Morley Mao, Ramesh Govindan, George
Varghese & Randy H. Katz, August 2002
p  “What is the sound of one route flapping?”
n  Tim Griffin, June 2002
p  Various work on routing convergence by Craig
Labovitz and Abha Ahuja a few years ago
p  “Happy Packets”
n  Closely related work by Randy Bush et al

Problem 1:
p  One path flaps:
n  BGP speakers pick next best path, announce to
all peers, flap counter incremented
n  Those peers see change in best path, flap
counter incremented
n  After a few hops, peers see multiple changes
simply caused by a single flap → prefix is
suppressed

Problem 2:
p  Different BGP implementations have
different transit time for prefixes
n  Some hold onto prefix for some time before
advertising
n  Others advertise immediately
p  Race to the finish line causes appearance
of flapping, caused by a simple
announcement or path change → prefix is
suppressed

Solution:
p  Misconfigured Route Flap Damping will seriously
impact access to:
n  Your network and
n  The Internet
p  More background contained in RIPE Routing
Working Group document:
n  www.ripe.net/ripe/docs/ripe-378
p  Recommendations now in:
n  www.rfc-editor.org/rfc/rfc7196.txt and www.ripe.net/
ripe/docs/ripe-580
130

Service Provider use of
Communities
Some examples of how ISPs
make life easier for themselves

BGP Communities
p  Another ISP “scaling technique”
p  Prefixes are grouped into different
“classes” or communities within the ISP
network
p  Each community means a different thing,
has a different result in the ISP network

BGP Communities
p  Communities are generally set at the edge
of the ISP network
n  Customer edge: customer prefixes belong to
different communities depending on the
services they have purchased
n  Internet edge: transit provider prefixes belong
to difference communities, depending on the
loadsharing or traffic engineering requirements
of the local ISP, or what the demands from its
BGP customers might be
p  Two simple examples follow to explain the
concept

Community Example:
Customer Edge
p  This demonstrates how communities
might be used at the customer edge of an
ISP network
p  ISP has three connections to the Internet:
n  IXP connection, for local peers
n  Private peering with a competing ISP in the
region
n  Transit provider, who provides visibility to the
entire Internet
p  Customers have the option of purchasing
combinations of the above connections

Community Example:
Customer Edge
p  Community assignments:
n  IXP connection: community 100:2100
n  Private peer: community 100:2200
p  Customer who buys local connectivity (via IXP) is
put in community 100:2100
p  Customer who buys peer connectivity is put in
community 100:2200
p  Customer who wants both IXP and peer
connectivity is put in 100:2100 and 100:2200
p  Customer who wants “the Internet” has no
community set
n  We are going to announce his prefix everywhere

CORE
Aggregation Router
CustomersCustomersCustomers
Community Example:
Customer Edge
p  Communities set at the aggregation router
where the prefix is injected into the ISP’s
iBGP
Border Router

Community Example:
Customer Edge
p  No need to alter filters at the network
border when adding a new customer
p  New customer simply is added to the
appropriate community
n  Border filters already in place take care of
announcements
n  ⇒ Ease of operation!

Community Example: Internet Edge
p  This demonstrates how communities might be
used at the peering edge of an ISP network
p  ISP has four types of BGP peers:
n  Customer
n  IXP peer
n  Private peer
n  Transit provider
p  The prefixes received from each can be classified
using communities
p  Customers can opt to receive any or all of the
above

p  Community assignments:
n  Customer prefix: community 100:3000
n  IXP prefix: community 100:3100
n  Private peer prefix: community 100:3200
p  BGP customer who buys local connectivity gets 100:3000
p  BGP customer who buys local and IXP connectivity receives
community 100:3000 and 100:3100
p  BGP customer who buys full peer connectivity receives
community 100:3000, 100:3100, and 100:3200
p  Customer who wants “the Internet” gets everything
n  Gets default route originated by aggregation router
n  Or pays money to get the full BGP table!

p  No need to create customised filters when
adding customers
n  Border router already sets communities
n  Installation engineers pick the appropriate
community set when establishing the customer
BGP session
n  ⇒ Ease of operation!

Community Example – Summary
p  Two examples of customer edge and
internet edge can be combined to form a
simple community solution for ISP prefix
policy control
p  More experienced operators tend to have
more sophisticated options available
n  Advice is to start with the easy examples
given, and then proceed onwards as
experience is gained

ISP BGP Communities
p  There are no recommended ISP BGP communities apart
from
n  RFC1998
n  The five standard communities
p  www.iana.org/assignments/bgp-well-known-communities
p  Efforts have been made to document from time to time
n  totem.info.ucl.ac.be/publications/papers-elec-versions/draft-
quoitin-bgp-comm-survey-00.pdf
n  But so far… nothing more… L
n  Collection of ISP communities at www.onesc.net/communities
n  NANOG Tutorial: www.nanog.org/meetings/nanog40/
presentations/BGPcommunities.pdf
p  ISP policy is usually published
n  On the ISP’s website
n  Referenced in the AS Object in the IRR

ISP Examples: Sprint
More info at
https://www.sprint.net/index.php?p=policy_bgp

ISP Example: NTT
More info at www.us.ntt.net/
about/policy/routing.cfm

ISP Example:
Verizon Europe
146
aut-num: AS702
descr: Verizon Business EMEA - Commercial IP service provider in Europe
<snip>
remarks: --------------------------------------------------------------
Verizon Business filters out inbound prefixes longer than /24.
We also filter any networks within AS702:RS-INBOUND-FILTER.
--------------------------------------------------------------
VzBi uses the following communities with its customers:
702:80 Set Local Pref 80 within AS702
702:120 Set Local Pref 120 within AS702
702:20 Announce only to VzBi AS'es and VzBi customers
702:30 Keep within Europe, don't announce to other VzBi AS's
702:1 Prepend AS702 once at edges of VzBi to Peers
702:2 Prepend AS702 twice at edges of VzBi to Peers
702:3 Prepend AS702 thrice at edges of VzBi to Peers
--------------------------------------------------------------
Advanced communities for customers
702:7020 Do not announce to AS702 peers with a scope of
National but advertise to Global Peers, European
Peers and VzBi customers.
702:7001 Prepend AS702 once at edges of VzBi to AS702
peers with a scope of National.
702:7002 Prepend AS702 twice at edges of VzBi to AS702
peers with a scope of National.
<snip>
And many more!

ISP Example:
Telia
147
aut-num: AS1299
descr: TeliaSonera International Carrier
<snip>
remarks: ------------------------------------------------------
remarks: BGP COMMUNITY SUPPORT FOR AS1299 TRANSIT CUSTOMERS:
remarks:
remarks: Community Action (default local pref 200)
remarks: -----------------------------------------
remarks: 1299:50 Set local pref 50 within AS1299 (lowest possible)
remarks: 1299:150 Set local pref 150 within AS1299 (equal to peer, backup)
remarks:
remarks: European peers
remarks: Community Action
remarks: --------- ------
remarks: 1299:200x All peers Europe incl:
remarks:
remarks: 1299:250x Sprint/1239
remarks: 1299:251x Savvis/3561
remarks: 1299:252x NTT/2914
remarks: 1299:253x Zayo/Abovenet/6461
remarks: 1299:254x FT/5511
remarks: 1299:255x GBLX/3549
remarks: 1299:256x Level3/3356
<snip>
remarks: Where x is number of prepends (x=0,1,2,3) or do NOT announce (x=9)
And many
many more!

ISP Example:
BT Ignite
148
aut-num: AS5400
descr: BT Ignite European Backbone
<snip>
remarks: The following BGP communities can be set by BT
remarks: BGP customers to affect announcements to major peers.
remarks:
remarks: 5400:NXXX
remarks: N=1 not announce
remarks: N=2 prepend an extra "5400 5400" on announcement
remarks: Valid values for XXX:
remarks: 000 All peers and transits
remarks: 500 All transits
remarks: 503 Level3 AS3356
remarks: 509 Telia AS1299
remarks: 510 NTT Verio AS2914
remarks: 002 Sprint AS1239
remarks: 003 Savvis AS3561
remarks: 004 C&W AS1273
remarks: 005 Verizon EMEA AS702
remarks: 014 DTAG AS3320
remarks: 016 Opentransit AS5511
remarks: 018 GlobeInternet Tata AS6453
remarks: 023 Tinet AS3257
remarks: 027 Telia AS1299
remarks: 045 Telecom Italia AS6762
remarks: 073 Eurorings AS286
remarks: 169 Cogent AS174
<snip>
And many
more!

ISP Example:
Level3
149
aut-num: AS3356
descr: Level 3 Communications
<snip>
remarks: --------------------------------------------------------
remarks: customer traffic engineering communities - Suppression
remarks: --------------------------------------------------------
remarks: 64960:XXX - announce to AS XXX if 65000:0
remarks: 65000:0 - announce to customers but not to peers
remarks: 65000:XXX - do not announce at peerings to AS XXX
remarks: --------------------------------------------------------
remarks: customer traffic engineering communities - Prepending
remarks: --------------------------------------------------------
remarks: 65001:0 - prepend once to all peers
remarks: 65001:XXX - prepend once at peerings to AS XXX
remarks: 65002:0 - prepend twice to all peers
remarks: 65002:XXX - prepend twice at peerings to AS XXX
<snip>
remarks: --------------------------------------------------------
remarks: customer traffic engineering communities - LocalPref
remarks: --------------------------------------------------------
remarks: 3356:70 - set local preference to 70
remarks: --------------------------------------------------------
remarks: customer traffic engineering communities - Blackhole
remarks: --------------------------------------------------------
remarks: 3356:9999 - blackhole (discard) traffic
<snip>
And many
more!

Deploying BGP in an ISP
Network
Okay, so we’ve learned all
about BGP now; how do we use
it on our network??

Deploying BGP
p  The role of IGPs and iBGP
p  Aggregation
p  Receiving Prefixes
p  Configuration Tips

The role of IGP and
iBGP
Ships in the night?
Or
Good foundations?

BGP versus OSPF/ISIS
p  Internal Routing Protocols (IGPs)
n  Examples are ISIS and OSPF
n  Used for carrying infrastructure addresses
n  NOT used for carrying Internet prefixes or
customer prefixes
n  Design goal is to minimise number of prefixes
in IGP to aid scalability and rapid convergence
154

p  BGP is used
n  Internally (iBGP)
n  Externally (eBGP)
p  iBGP is used to carry:
n  Some/all Internet prefixes across backbone
n  Customer prefixes
p  eBGP is used to:
n  Exchange prefixes with other ASes
n  Implement routing policy
155

156
BGP/IGP model used in ISP
networks
p  Model representation
IGP
iBGP
eBGP eBGP eBGP
AS1 AS2 AS3 AS4
IGP
iBGP
IGP
iBGP
IGP
iBGP

p  DO NOT:
n  Distribute BGP prefixes into an IGP
n  Distribute IGP routes into BGP
n  Use an IGP to carry customer prefixes
p  YOUR NETWORK WILL NOT SCALE

Injecting prefixes into iBGP
p  Use iBGP to carry customer prefixes
n  Don’t ever use IGP
p  Point static route to customer interface
p  Enter network into BGP process
n  Ensure that implementation options are used
so that the prefix always remains in iBGP,
regardless of state of interface
n  i.e. avoid iBGP flaps caused by interface flaps

Aggregation
Quality or Quantity?

Aggregation
p  Aggregation means announcing the address block received
from the RIR to the other ASes connected to your network
p  Subprefixes of this aggregate may be:
n  Used internally in the ISP network
n  Announced to other ASes to aid with multihoming
p  Too many operators are still thinking about class Cs,
resulting in a proliferation of /24s in the Internet routing
table
n  January 2016: 318000 /24s in IPv4 table of 580000 prefixes
p  The same is happening for /48s with IPv6
n  January 2016: 11800 /48s in IPv6 table of 25800 prefixes
160

Aggregation – Example
161
p  Customer has /23 network assigned from
AS100’s /19 address block
p  AS100 announces customers’ individual
networks to the Internet
AS100
customer
100.10.10.0/23Internet
100.10.10.0/23
100.10.0.0/24
100.10.4.0/22
…

Aggregation – Bad Example
p  Customer link goes down
n  Their /23 network
becomes unreachable
n  /23 is withdrawn from
AS100’s iBGP
p  Their ISP doesn’t
aggregate its /19 network
block
n  /23 network withdrawal
announced to peers
n  starts rippling through the
Internet
n  added load on all Internet
backbone routers as
network is removed from
routing table
p  Customer link returns
n  Their /23 network is now
visible to their ISP
n  Their /23 network is re-
advertised to peers
n  Starts rippling through
Internet
n  Load on Internet backbone
routers as network is
reinserted into routing
table
n  Some ISP’s suppress the
flaps
n  Internet may take 10-20
min or longer to be visible
n  Where is the Quality of
Service???

Aggregation – Example
163
p  Customer has /23 network assigned from
AS100’s /19 address block
p  AS100 announced /19 aggregate to the
Internet
AS100
customer
100.10.10.0/23
100.10.0.0/19
aggregate
Internet
100.10.0.0/19

p  Customer link returns
p  Their /23 network is
visible again
n  The /23 is re-injected
into AS100’s iBGP
p  The whole Internet
becomes visible
immediately
p  Customer has Quality
of Service perception
p  Customer link goes
down
n  their /23 network
becomes unreachable
n  /23 is withdrawn from
AS100’s iBGP
p  /19 aggregate is still
being announced
n  no BGP hold down
problems
n  no BGP propagation
delays
n  no damping by other
ISPs
Aggregation – Good Example

Aggregation – Summary
p  Good example is what everyone should
do!
n  Adds to Internet stability
n  Reduces size of routing table
n  Reduces routing churn
n  Improves Internet QoS for everyone
p  Bad example is what too many still do!
n  Why? Lack of knowledge?
n  Laziness?

Separation of iBGP and eBGP
p  Many ISPs do not understand the importance of
separating iBGP and eBGP
n  iBGP is where all customer prefixes are carried
n  eBGP is used for announcing aggregate to Internet and
for Traffic Engineering
p  Do NOT do traffic engineering with customer
originated iBGP prefixes
n  Leads to instability similar to that mentioned in the
earlier bad example
n  Even though aggregate is announced, a flapping
subprefix will lead to instability for the customer
concerned
p  Generate traffic engineering prefixes on the
Border Router

The Internet Today
(January 2016)
p  Current Internet Routing Table Statistics
n  BGP Routing Table Entries 579519
n  Prefixes after maximum aggregation 213882
n  Unique prefixes in Internet 282120
n  Prefixes smaller than registry alloc 189985
n  /24s announced 317953
n  ASes in use 52493
n  (maximum aggregation is calculated by Origin AS)
n  (unique prefixes > max aggregation means that
operators are announcing aggregates from their blocks
without a covering aggregate)
167

Efforts to improve aggregation
p  The CIDR Report
n  Initiated and operated for many years by Tony Bates
n  Now combined with Geoff Huston’s routing analysis
p  www.cidr-report.org
p  (covers both IPv4 and IPv6 BGP tables)
n  Results e-mailed on a weekly basis to most operations
lists around the world
n  Lists the top 30 service providers who could do better at
aggregating
p  RIPE Routing WG aggregation recommendations
n  IPv4: RIPE-399 — www.ripe.net/ripe/docs/ripe-399.html
n  IPv6: RIPE-532 — www.ripe.net/ripe/docs/ripe-532.html
168

Efforts to Improve Aggregation
The CIDR Report
p  Also computes the size of the routing table
assuming ISPs performed optimal aggregation
p  Website allows searches and computations of
aggregation to be made on a per AS basis
n  Flexible and powerful tool to aid ISPs
n  Intended to show how greater efficiency in terms of BGP
table size can be obtained without loss of routing and
policy information
n  Shows what forms of origin AS aggregation could be
performed and the potential benefit of such actions to
the total table size
n  Very effectively challenges the traffic engineering excuse

BGP Techniques for Network Operators

Importance of Aggregation
p  Size of routing table
n  Router Memory is not so much of a problem as it was in
the 1990s
n  Routers can be specified to carry 1 million+ prefixes
p  Convergence of the Routing System
n  This is a problem
n  Bigger table takes longer for CPU to process
n  BGP updates take longer to deal with
n  BGP Instability Report tracks routing system update
activity
n  http://bgpupdates.potaroo.net/instability/bgpupd.html

Aggregation Potential
(source: bgp.potaroo.net)
AS Path
AS Origin

Aggregation
Summary
p  Aggregation on the Internet could be
MUCH better
n  35% saving on Internet routing table size is
quite feasible
n  Tools are available
n  Commands on the routers are not hard
n  CIDR-Report webpage

Receiving Prefixes
p  There are three scenarios for receiving
prefixes from other ASNs
n  Customer talking BGP
n  Peer talking BGP
n  Upstream/Transit talking BGP
p  Each has different filtering requirements
and need to be considered separately

Receiving Prefixes:
From Customers
p  ISPs should only accept prefixes which have been
assigned or allocated to their downstream
customer
p  If ISP has assigned address space to its
customer, then the customer IS entitled to
announce it back to his ISP
p  If the ISP has NOT assigned address space to its
customer, then:
n  Check the five RIR databases to see if this address
space really has been assigned to the customer
n  The tool: whois –h jwhois.apnic.net x.x.x.0/24
p  (jwhois queries all RIR databases)

Receiving Prefixes:
From Customers
p  Example use of whois to check if customer is
entitled to announce address space:
$ whois -h jwhois.apnic.net 202.12.29.0
inetnum: 202.12.28.0 - 202.12.29.255
netname: APNIC-AP
descr: Asia Pacific Network Information Centre
descr: Regional Internet Registry for the Asia-Pacific
descr: 6 Cordelia Street
descr: South Brisbane, QLD 4101
descr: Australia
country: AU
admin-c: AIC1-AP
tech-c: NO4-AP
mnt-by: APNIC-HM
mnt-irt: IRT-APNIC-AP
changed: hm-changed@apnic.net
status: ASSIGNED PORTABLE
changed: hm-changed@apnic.net 20110309
source: APNIC
184
Portable – means its an
assignment to the customer, the
customer can announce it to you

Receiving Prefixes:
From Customers
p  Example use of whois to check if customer is entitled
to announce address space:
$ whois -h whois.ripe.net 193.128.0.0
inetnum: 193.128.0.0 - 193.133.255.255
netname: UK-PIPEX-193-128-133
descr: Verizon UK Limited
country: GB
org: ORG-UA24-RIPE
admin-c: WERT1-RIPE
tech-c: UPHM1-RIPE
status: ALLOCATED UNSPECIFIED
remarks: Please send abuse notification to abuse@uk.uu.net
mnt-by: RIPE-NCC-HM-MNT
mnt-lower: AS1849-MNT
mnt-routes: AS1849-MNT
mnt-routes: WCOM-EMEA-RICE-MNT
mnt-irt: IRT-MCI-GB
source: RIPE # Filtered
185
ALLOCATED – means that this is
Provider Aggregatable address
space and can only be announced
by the ISP holding the allocation
(in this case Verizon UK)

Receiving Prefixes:
From Peers
p  A peer is an ISP with whom you agree to
exchange prefixes you originate into the
Internet routing table
n  Prefixes you accept from a peer are only those
they have indicated they will announce
n  Prefixes you announce to your peer are only
those you have indicated you will announce

Receiving Prefixes:
From Peers
p  Agreeing what each will announce to the
other:
n  Exchange of e-mail documentation as part of
the peering agreement, and then ongoing
updates
OR
n  Use of the Internet Routing Registry and
configuration tools such as the IRRToolSet
https://github.com/irrtoolset/irrtoolset
187

Receiving Prefixes:
From Upstream/Transit Provider
p  Upstream/Transit Provider is an ISP who you pay
to give you transit to the WHOLE Internet
p  Receiving prefixes from them is not desirable
unless really necessary
n  Traffic Engineering – see BGP Multihoming Presentations
p  Ask upstream/transit provider to either:
n  originate a default-route
OR
n  announce one prefix you can use as default

Receiving Prefixes:
p  If necessary to receive prefixes from any
provider, care is required.
n  Don’t accept default (unless you need it)
n  Don’t accept your own prefixes
p  Special uses prefixes for IPv4 and IPv6:
n  http://www.rfc-editor.org/rfc/rfc6890.txt
p  For IPv4:
n  Don’t accept prefixes longer than /24 (?)
p  /24 was the historical class C
p  For IPv6:
n  Don’t accept prefixes longer than /48 (?)
p  /48 is the design minimum delegated to a site
189

Receiving Prefixes:
p  Check Team Cymru’s list of “bogons”
www.team-cymru.org/Services/Bogons/http.html
p  For IPv4 also consult:
www.rfc-editor.org/rfc/rfc6441.txt (BCP171)
p  For IPv6 also consult:
www.space.net/~gert/RIPE/ipv6-filters.html
p  Bogon Route Server:
www.team-cymru.org/Services/Bogons/routeserver.html
n  Supplies a BGP feed (IPv4 and/or IPv6) of address blocks
which should not appear in the BGP table
190

Receiving IPv4 Prefixes
191
deny 0.0.0.0/0 ! Default
deny 0.0.0.0/8 to /32 ! RFC1122 local host
deny 10.0.0.0/8 to /32 ! RFC1918
deny 100.64.0.0/10 to /32 ! RFC6598 shared address
deny 127.0.0.0/8 to /32 ! Loopback
deny 169.254.0.0/16 to /32 ! Auto-config
deny 172.16.0.0/12 to /32 ! RFC1918
deny 192.0.0.0/24 to /32 ! RFC6598 IETF protocol
deny 192.0.2.0/24 to /32 ! TEST1
deny 192.168.0.0/16 to /32 ! RFC1918
deny 198.18.0.0/15 to /32 ! Benchmarking
deny 198.51.100.0/24 to /32 ! TEST2
deny 203.0.113.0/24 to /32 ! TEST3
deny 224.0.0.0/3 to /32 ! Multicast & Experimental
deny 0.0.0.0/0 from /25 to /32 ! Prefixes >/24
deny subnets of your own address space
permit everything else

Receiving IPv6 Prefixes
192
permit 64:ff9b::/96 ! RFC6052 v4v6trans
permit 2001::/32 ! Teredo
deny 2001::/23 to /128 ! RFC2928 IETF protocol
deny 2001:2::/48 to /128 ! Benchmarking
deny 2001:10::/28 to /128 ! ORCHID
deny 2001:db8::/32 to /128 ! Documentation
permit 2002::/16 ! 6to4 aggregate
deny 2002::/16 to /128 ! 6to4 subnets
deny 3ffe::/16 to /128 ! Old 6bone
deny subnets of your own address block
permit 2000::/3 to /48 ! Global Unicast to /48s
deny ::/0 to /128 ! Deny everything else

Receiving Prefixes
p  Paying attention to prefixes received from
customers, peers and transit providers
assists with:
n  The integrity of the local network
n  The integrity of the Internet
p  Responsibility of all ISPs to be good
Internet citizens

Configuration Tips
Of passwords, tricks and
templates

iBGP and IGPs
Reminder!
p  Make sure loopback is configured on
router
n  iBGP between loopbacks, NOT real interfaces
p  Make sure IGP carries loopback IPv4 /32
and IPv6 /128 address
p  Consider the DMZ nets:
n  Use unnumbered interfaces?
n  Use next-hop-self on iBGP neighbours
n  Or carry the DMZ IPv4 /30s and IPv6 /127s in
the iBGP
n  Basically keep the DMZ nets out of the IGP!

iBGP: Next-hop-self
p  BGP speaker announces external network
to iBGP peers using router’s local address
(loopback) as next-hop
p  Used by many ISPs on edge routers
n  Preferable to carrying DMZ point-to-point
addresses in the IGP
n  Reduces size of IGP to just core infrastructure
n  Alternative to using unnumbered interfaces
n  Helps scale network
n  Many ISPs consider this “best practice”

Limiting AS Path Length
p  Some BGP implementations have
problems with long AS_PATHS
n  Memory corruption
n  Memory fragmentation
p  Even using AS_PATH prepends, it is not
normal to see more than 20 ASes in a
typical AS_PATH in the Internet today
n  The Internet is around 5 ASes deep on average
n  Largest AS_PATH is usually 16-20 ASNs

p  Some announcements have ridiculous lengths of AS-paths
n  This example is an error in one IPv6 implementation
n  This example shows 100 prepends (for no obvious reason)
p  If your implementation supports it, consider limiting the
maximum AS-path length you will accept
*> 3FFE:1600::/24 22 11537 145 12199 10318 10566 13193 1930 2200
3425 293 5609 5430 13285 6939 14277 1849 33 15589 25336 6830 8002 2042
7610 i
*>i193.105.15.0 2516 3257 50404 50404 50404 50404 50404 50404
50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404
50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404
50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404
50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404
50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404
50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404
50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404
50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 50404 i
Limiting AS Path Length

BGP Maximum Prefix Tracking
p  Allow configuration of the maximum number of
prefixes a BGP router will receive from a peer
n  Supported by good BGP implementations
p  Usually have two level control for prefix count:
n  Reaches warning threshold: log a warning message
p  Threshold is configurable
n  Reaches maximum:
p  Only send warnings
p  Tear down BGP, manual intervention required to restart
p  Tear down BPG and automatically restart after a delay
(configurable)

BGP TTL “hack”
p  Implement RFC5082 on BGP peerings
n  (Generalised TTL Security Mechanism)
n  Neighbour sets TTL to 255
n  Local router expects TTL of incoming BGP packets to be
254
n  No one apart from directly attached devices can send
BGP packets which arrive with TTL of 254, so any
possible attack by a remote miscreant is dropped due to
TTL mismatch
ISP AS 100
Attacker
TTL 254
TTL 253 TTL 254
R1 R2

BGP TTL “hack”
p  TTL Hack:
n  Both neighbours must agree to use the feature
n  TTL check is much easier to perform than MD5
n  (Called BTSH – BGP TTL Security Hack)
p  Provides “security” for BGP sessions
n  In addition to packet filters of course
n  MD5 should still be used for messages which slip
through the TTL hack
n  See https://www.nanog.org/meetings/nanog27/
presentations/meyer.pdf for more details

Templates
p  Good practice to configure templates for
everything
n  Vendor defaults tend not to be optimal or even
very useful for ISPs
n  ISPs create their own defaults by using
configuration templates
p  eBGP and iBGP examples follow
n  Also see Team Cymru’s BGP templates
p  http://www.team-cymru.org/ReadingRoom/
Documents/

iBGP Template
Example
p  iBGP between loopbacks!
p  Next-hop-self
n  Keep DMZ and external point-to-point out of IGP
p  Always send communities in iBGP
n  Otherwise accidents will happen
n  (Default on some vendor implementations, optional on
others)
p  Hardwire BGP to version 4
n  Yes, this is being paranoid!
n  Prevents accidental configuration of version 3 BGP still
supported in some implementations

iBGP Template
Example continued
p  Use passwords on iBGP session
n  Not being paranoid, VERY necessary
n  It’s a secret shared between you and your peer
n  If arriving packets don’t have the correct MD5 hash,
they are ignored
n  Helps defeat miscreants who wish to attack BGP
sessions
p  Powerful preventative tool, especially when
combined with filters and the TTL “hack”

eBGP Template
Example
p  BGP damping
n  Do NOT use it unless you understand the impact
n  Do NOT use the vendor defaults without thinking
p  Remove private ASes from announcements
n  Common omission today
p  Use extensive filters, with “backup”
n  Use as-path filters to backup prefix filters
n  Keep policy language for implementing policy, rather
than basic filtering
p  Use password agreed between you and peer on
eBGP session

eBGP Template
Example continued
p  Use maximum-prefix tracking
n  Router will warn you if there are sudden increases in
BGP table size, bringing down eBGP if desired
p  Limit maximum as-path length inbound
p  Log changes of neighbour state
n  …and monitor those logs!
p  Make BGP admin distance higher than that of any
IGP
n  Otherwise prefixes heard from outside your network
could override your IGP!!

Summary
p  Use configuration templates
p  Standardise the configuration
p  Be aware of standard “tricks” to avoid
compromise of the BGP session
p  Anything to make your life easier, network less
prone to errors, network more likely to scale
p  It’s all about scaling – if your network won’t
scale, then it won’t be successful

BGP Techniques for Network Operators

More Related Content

What's hot (20)

Similar to BGP Techniques for Network Operators (20)

More from APNIC (20)

Recently uploaded (20)

BGP Techniques for Network Operators