This lecture is largely based on a BGP tutorial by T. Griffin from AT&T Research.
© J. Liebeherr, All rights reserved
1

Regional
Network
(Tier 2) local ISP
(Tier 3)
Backbone Network
(Tier 1)
IXP
Regional
Network
(Tier 2) local ISP
(Tier 3)
Backbone Network
(Tier 1)
IXP local ISP
(Tier 3)
Regional
Network
(Tier 2)
IXP corporate network
Regional
Network
(Tier 2) campus network
2

• Location where a network (ISP, corporate network, or regional network) gets access to the Internet is called a Point-of-
Presence (POP).
• Locations where Tier-1 or Tier-2 networks exchange traffic are called peering points .
– Public peering : Traffic is swapped in a specific location, called Internet exchange points (IXPs)
– Private peering: Two networks establish a direct link to each other.
3

• Outside: • Inside:
4

5

6

• An autonomous system (AS) is a region of the Internet that is administered by a single entity and that has a unified routing policy
• Each autonomous system is assigned an Autonomous System Number
(ASN).
• UofT’s campus network (AS239)
• Rogers Cable Inc. (AS812)
• Sprint (AS1239, AS1240, AS 6211, …)
• Interdomain routing is concerned with determining paths between autonomous systems (interdomain routing )
• Routing protocols for interdomain routing are called exterior gateway protocols (EGP)
7

Review: Interdomain and Intradomain Routing
AS 2 AS 5
AS 1
AS 6
AS 7
AS 3
AS 4
• Routing protocols for intradomain routing are called interior gateway protocols (IGP)
– Objective: shortest path
• Routing protocols for interdomain routing are called exterior gateway protocols (EGP)
– Objective: satisfy policy of the AS
8

9

AS 2
IGP (e.g., OSPF)
EGP (e.g., BGP)
• Interior Gateway Protocol
– Routing is done based on metrics
– Routing domain is one autonomous system
• Exterior Gateway Protocol
– Routing is done based on policies
– Routing domain is the entire Internet
AS 2
IGP (e.g., RIP)
10

• Interdomain routing is based on connectivity between autonomous systems
• Interdomain routing can ignore many details of router interconnection
AS 1 AS 2
AS 3
11

AT&T North America
From: T. Griffin, BGP Tutorial
12

• Multiple routing protocols can run on the same router
• Each routing protocol updates the routing table routing protocol
RIP
Process
BGP
Process
OSPF
Process routing table updates routing protocol incoming IP datagrams routing table routing table lookup
IP
Forwarding outgoing IP datagrams
13

• local traffic
• transit traffic
• Stub AS
= traffic with source or destination in AS
= traffic that passes through the AS
= has connection to only one AS, only carry local traffic
• Multihomed Stub AS = has connection to >1 AS, but does not carry transit traffic
• Transit AS = has connection to >1 AS and carries transit traffic
14

AS 1
AS 3
Settings:
• AS 1 is a multi-homed stub network
• AS 3 and AS 4 are transit networks
• AS 2 and AS 5 are is a stub networks
AS 5
AS 2
AS 4
15

Example:
• AS 3 carries traffic between AS 1 and AS 4 and between AS 2 and AS 4
• But AS 3 does not carry traffic between AS 1 and AS 2
AS 1
• The example shows a routing policy.
AS 3
AS 4
AS 2
16

Customer/
Provider
AS 2
Customer/
Provider
AS 4
Customer/
Provider
AS 6
Customer/
Provider
AS 7
AS 5
Customer/
Provider
AS 8
• A stub network typically obtains access to the Internet through a transit network.
• Transit network that is a provider may be a customer for another network
• Customer pays provider for service
17

AS 1
Customer/
Provider
Peers
AS 2
Customer/
Provider
Peers
AS 4 AS 5
Customer/Provider
Customer/
Provider
AS 7
AS 8
• Transit networks can have a peer relationship
• Peers provide transit between their respective customers
• Peers do not provide transit between peers
• Peers normally do not pay each other for service
AS 3
Customer/
Provider
AS 6
18

AS 1
Customer/
Provider
Peers
AS 2
Customer/
Provider
AS 4
Peers
Customer/Provider
AS 5
Peers
Customer/
Provider
AS 7
AS 8
• Note that peering reduces upstream traffic
• Delays can be reduced through peering
• But: Peering may not generate revenue
AS 3
Customer/
Provider
AS 6
19

• Border Gateway Protocol is the interdomain routing protocol for the Internet for routing between autonomous systems
• Currently in version 4 (1995)
– Network administrators can specify routing policies
– BGP is a path vector protocol (Like distance vector, but routing messages in BGP contain complete routes)
• Uses TCP to transmit routing messages
20

• An autonomous system uses BGP to advertise its network address(es) to other AS’s
• BGP helps an AS to:
1.
Learn about reachable networks from neighboring AS’s
2. Distribute the information about reachable networks to routers inside the AS
3. Select a route if there are multiple routes to reach the same network
21

• Open: Establishes a peering session
• Notification: Closes a peering session
• Keep Alive: Handshake at regular intervals to maintain peering session
• Update: Announces new routes or withdraws previously announced routes.
Each announced route is specified as a network prefix with attribute values
22

AS 1
• The networks that are advertised are network IP addresses with a prefix, E.g., 128.100.0.0/16
Prefixes reachable from AS 1
AS 2
Prefixes reachable from AS 3
AS 3
23

• BGP is executed between two routers
– BGP session
– BGP peers or BGP speakers
• Procedure:
1. Establishes TCP connection (port 175) to BGP peer
2. Exchange all BGP routes
3. As long as connection is alive:
Periodically send incremental updates
• Note: Not all autonomous systems need to run BGP. On many stub networks, the route to the provider can be statically configured
AS 1
AS 2
BGP Session
24

• BGP peers advertise reachability of IP networks
• A advertises a path to a network
(e.g., 10.0.0.0/8) to B only if it is willing to forward traffic going to that network
BGP Peer
Advertise path to 10.0.0.0/24
A
B
BGP Peer
• Path-Vector:
– A advertises the complete path AS A, …., AS X
 this avoids loops
10.0.0.0/24
25

• External BGP session (eBGP):
Peers are in different AS’es
• Internal BGP session (iBGP)
Peers are in the same AS
• Note that iBGP sessions use routes constructed by an intradomain routing protocol to exchange messages !
AS B eBGP session
AS A iBGP session
26

• All iBGP peers in the same autonomous system are fully meshed
• Peer announces routes received via eBGP to iBGP peers
• But: iBGP peers do not announce routes received via iBGP to other iBGP peers
Update from eBGP session
AS A
27

• Full mesh of iBGP routers is difficult to maintain
• Router Reflectors (RR) present an alternative
• All iBGP routers peer with the RR
– RR acts as a server
– Other iBGP routers become clients
Update from eBGP session
AS A
RR
28

• A BGP routers route advertisement is sent in a BGP UPDATE message
• A route is announced as a Network Prefix and Attributes
• Attributes specify details about a route:
– Mandatory attributes:
ORIGIN
AS_PATH
NEXT_HOP
– many other attributes
29

• Originating domain sends a route with ORIGIN attribute
10.0.1.0/24,
ORIGIN {1}
10.0.1.0/24,
ORIGIN {1}
10.0.1.0/24,
ORIGIN {1}
10.0.1.0/24,
ORIGIN {1}
10.0.1.0/24,
ORIGIN {1}
30

• Each AS that propagates a route prepends its own AS number
– AS-PATH collects a path to reach the network prefix
• Path information prevents routing loops from occuring
• Path information also provides information on the length of a path (By default, a shorter route is preferred)
• Note: BGP aggregates routes according to CIDR rules
AS 2 AS 4
10.0.1.0/24,
AS-PATH {4,2,1}
10.0.1.0/24,
AS-PATH {1}
10.0.1.0/24,
AS-PATH {2,1}
AS 1 AS 5
10.0.1.0/24,
AS-PATH {1}
AS 3
10.0.1.0/24,
AS-PATH {3,1}
31

• Each router that sends a route advertisement it includes its own IP address in a NEXT-HOP attribute
• The attribute provides information for the routing table of the receiving router.
AS 1
128.100.11.1
128.143.71.21
AS 5
AS 3
10.0.1.0/24,
NEXT-HOP {128.100.11.1}
10.0.1.0/24,
NEXT-HOP {128.143.71.21}
32

192.0.1.2
AS 1
128.100.11.1/24 eBGP
10.0.1.0/24,
NEXT-HOP {128.100.11.1}
At R1:
Routing table
Dest.
Next hop
128.100.11.0/24 192.0.1.2
BGP info
Dest.
10.0.1.0/24
Next hop
128.100.11.1
IGP router
AS 3 iBGP
10.0.1.0/24,
NEXT-HOP {128.100.11.1}
R1
Routing table
Dest.
Next hop
128.100.11.0/24 192.0.1.2
10.0.1.0/24 192.0.1.2
33

• An AS may get more than one route to an address
• Needs to select a route
AS 1
Advertise path to
10.0.1.0/24
AS 2
AS 1
Advertise path to
10.0.1.0/24
Advertise path to
10.0.1.0/24
AS 3 Advertise path to
10.0.1.0/24
Route Selection Criteria (in order of preference)
• Highest Local Preference
• Shortest AS-Path
• Lowest MED (multi-exit discriminator) (  called “metric” in BGP)
• Prefer iBGP over eBGP routes
• Lowest IGP cost to leave AS (“hot potato”)
• Lowest router ID (  used as tie breaker)
AS 4
34

AS 1
AS 1
Advertise path to
10.0.1.0/24
AS 2
AS 2
Local pref = 10
Local pref
= 100
Advertise path to
10.0.1.0/24
AS 1
AS 1
Local pref = 80
Advertise path to
10.0.1.0/24
AS 3
AS 3
Local pref = 50
Advertise path to
AS 4
10.0.1.0/24
AS 4
• If there are multiple exit points from the AS, the local preference attribute is used to select the exit point for a specific route
• Local Preference is used only for iBGP sessions
• Value is set locally
35

• Router R3 in autonomous system
A receives two advertisements to
AS A
– Which route should it pick?
• Hot Potato Rule: Select the iBGP peer that has the shortest IGP route
• Analogy: Get the packet out of one’s own AS as quickly as possible, i.e., on the shortest path
Route to X
Route to X
R1 R2
Route to X Route to X
R3
AS A
36

Finding the cheapest IGP route:
• Compare the cost of the two paths
– R3  R1
– R3  R2 according to the IGP protocol
• Here: R1 has the shortest path
• Add a routing table entry for destination X
Route to X
Route to X
R1
Cost=6
Cost=23
R2
R3
AS A
37

• AS1 would serve its customer (source) better by not picking the shortest route to AS 2
• In fact, customer may have paid for a high-bandwidth service!
Source
AS 1
Cost=20
High bandwidth network
Cost=5
Low bandwidth network
AS 2
Destination
38

BGP updates arrive
Filter routes and change attributes
Based on attributes
Best entry is entered in
IP routing table
Filter routes and change attributes
BGP updates arrive
Apply Import
Rules
Select Best
Route
Update IP routing table
Apply Export
Rules
IP routing table
39

• An AS may not accept all routes that are advertised
• An AS may not advertise certain routes
• Route policies determines which routes are filtered
• If an AS wants to have less inbound traffic it should adapt its export rules
• If an AS wants to control its outbound traffic, it adapts its import rules
Control
Inbound traffic
AS A
Control
Outbound traffic
Change export rules
Change import rules
40

• Since AS 5 is a stub network it should not advertise routes to networks other than networks in AS 5
• When AS 3 learns about the path {AS1,
AS4}, it should not advertise the route
{AS3, AS1, AS4} to
AS 2.
AS 4 AS 1
P ee rs
AS 3
P ee rs
Customer/
Provider
AS 2
Peers
Customer/Provider
Customer/Provider
AS 6
AS 5
41

• In many cases, packets are routed according to the AS-PATH
AS 1
128.100.0.0/16
AS 2
128.100.0.0/16,
AS-PATH {3,2,1}
AS 3 AS 5
• However, in some cases this is not true
(Here: AS 2 filters routes with a long prefix)
128.100.0.0/16,
AS-PATH {1}
AS 1
128.100.0.0/16
128.100.22.0/24,
AS-PATH {4}
AS 2
128.100.0.0/16,
AS-PATH {2, 1}
128.100.0.0/16,
AS-PATH {3,2,1}
AS 3
Does not advertise /24 networks
AS 5
AS 4
128.100.22.0/24
42

Short AS-PATH does not mean that route is short
• From AS 6’s perspective
– Path {AS2, AS1} is short
– Path {AS5, AS4, AS3, AS1} is long
AS 3
• But the number of traversed routers is larger when using the shorter AS-PATH
AS 4
AS 1
AS 2
AS 5
AS 6
43

Source: Geoff Huston. http://www.telstra.net/ops/bgptable.html on August 8, 2001
44

Source: bgp.potaroo.net, 2010
45

• BGP is a simple protocol but it is very difficult to configure
• BGP has severe stability issue due to policies  BGP is known to not converge
• As of July 2005, 39,000 AS numbers (of available 64,510) are consumed
46