Technical Aspects of Peering

1
Technical Aspects of Peering

• Session 4

2
Overview

• Peering checklist/requirements
• Peering step by step
• Peering arrangements and options
• Exercises

3
Peering requirements

• BGP V4 router with enough memory to receive all
  routes
• 256MB minimum for the global routing table
• 32MB for all African routes
• Less for just your country
• Portable address space (ie not received from your
  upstream ISPS this space is already being
  advertised to peers upstream)
• AS Number
• Both obtainable from AfriNIC

4
Additional requirements

• List of prefixes that will be advertised and
  received from peers
• IP Address of each peer (including your own
  border router)
• Maximum number of hops between the BGP routers if
  they are not adjacent to each other
• eBGP multihop is not recommended, and not
  included in the handout

5
Peering step by step Step 1

• John Doe Communications wants to peer with Expert
  Networks.
• Step 1 Write down all the necessary information
  for each party
• A
• Company Name John Doe Communications
• AS number AS100
• Address space 12.1.1.0/24, 196.25.0.0/16
• Border router Cisco 2621
• BGP peer address 12.1.1.10
• B
• Company Name Expert Networks
• AS number AS200
• Address space 150.200.54.0/23
• Border router Linux PC running Quagga
• BGP peer address 150.200.54.125

A
6
Step 2.1

• Configure a loopback interface on the router.
  This is necessary in order to have an BGP peer
  with an interface that will always be up even if
  some of the physical interfaces on the router go
  down. (Especially useful with iBGP.)
• interface Loopback0
• ip address 12.1.1.10 255.255.255.255

7
Step 2.3

• Define filters to advertise and receive only the
  routes we know about. This is very important. If
  this is omitted any peer can flood your routing
  table with bogus entries. It can also cause your
  router to crash if too many prefixes are accepted
  by your router.
• ! accept all prefixes smaller or equal to /24,
• ! but only from the address space that we know
• ! belongs to each AS. AS 100 is our own AS.
• ip prefix-list AS100 seq 5 permit 12.1.1.0/24
  
• ip prefix-list AS100 seq 10 permit 196.25.0.0/16
  le 24
• ! AS200 is our peer
• ip prefix-list AS200 seq 5 permit
  150.200.54.0/23 le 24

8
Step 2.4

• The rest of the settings all reside in the BGP
  section of the configuration. Specify your AS
  number here.
• ! configure BGP sessions
• router BGP 100
• By default BGP does not advertise a route until
  all routes within the AS have learned of the
  route through the IGP. This command enables BGP
  to advertise routes to peers without
  synchronizing the route to the IGP.
• no synchronization

9
Step 2.5

• Log all changes such as BGP connections going
  down. These changes can be monitored by exporting
  the router logs to a syslog server. Most ISPs
  have a central log server and have technicians
  monitoring all events.
• bgp log-neighbour-changes
• Do not use redistribute commands to get routes
  into BGP. They make it too easy for unwanted
  routes to appear in your BGP tables.

10
Step 2.5

• Add a network statement for each route that you
  will announce. Also add a null route for
  aggregates that might not already be in your IGP
  routing table. Without these commands no routes
  in our routing table would be advertised to any
  peers.
• ! ensure that the aggregate route is always
  present
• ip route 196.25.0.0 255.255.0.0 null0 254
• ! add your own networks to BGP
• router bgp 100
• network 12.1.1.0 mask 255.255.255.0
• network 196.25.0.0 mask 255.255.0.0
• Do the above on only one router, or only a few
  routers in your AS, not on every router.

11
Step 2.6

• Do not attempt to summarize routes. This command
  is necessary if we want to exchange classless
  routes (i.e. routes other than Class A, B, or C
  routes).
• no auto-summary
• We now set up a peering session with Expert
  Networks (AS 200). If there was more than one
  peer we would have entered similar commands for
  each peer. The first command specifies the AS
  number of the peer (also known as a neighbour).
  
• neighbour 1.2.3.4 remote-as 200

12
Step 2.6

• Add a description. If there are many neighbours
  defined, it is useful to find the appropriate
  neighbour when configuration changes have to be
  made by looking at these descriptions.
• neighbour 1.2.3.4 description Expert Networks

13
Step 2.7

• This command instructs the router to set the
  gateways for all routes added to the routing
  table to itself. Always enable this when peering
  with other autonomous systems.
• neighbour 1.2.3.4 next-hop-self

14
Step 2.8

• Instruct the router to store received updates.
  This allows us to update a BGP session without
  having to restart the session.
• neighbour 1.2.3.4 soft-reconfiguration inbound
• This uses extra memory. In IOS 12 or later, you
  can get the same benefits using the BGP route
  refresh capability instead of using memory. Use
  show ip bgp neighbor x.x.x.x

15
Step 2.8

• Only advertise and accept routes allowed by our
  filters to prevent flooding of our routing table.
• neighbour 1.2.3.4 prefix-list AS100 out
• neighbour 1.2.3.4 prefix-list AS200 in

16
Step 3 Verify

• The following commands can be used to diagnose
  problems with your BGP configuration
• ! show a summary of peering sessions
• show ip bgp summary
• ! show neighbour details
• show ip bgp neighbours
• ! show routes received from neighbours
• show ip bgp
• ! show routes received from neighbour
  192.168.4.1
• show ip bgp neighbours 192.168.4.1
  received-routes
• show ip bgp neighbours 192.168.4.1 routes
• ! show routes advertised to neighbour
  192.168.4.3
• show ip bgp neighbours 192.168.4.3
  advertised-routes
• ! show all routes known via all protocols
• show ip route

17
Option 1 Mandatory multi-lateral peering

• All IXP participants peer with a central route
  server. This forces all to peer with all and
  reduces the number of peering sessions that has
  to be maintained by each peer.
• A central route server ! a route reflector!!
  Route reflectors used within iBGP to eliminate
  the need for a fully meshed network.

18
Option 1 Mandatory multi-lateral peering

• Advantages
• Automatic peering with all - easy
• Complexity is centralised easy for ISPs
• Easy to connect only one BGP session

• Disadvantages
• Forced peering with all - inflexible
• Complexity is centralised hard for IXP operator
• Complex policies are impossible

19
Option 2 Bilateral peering

• Option 1 doesnt scale well Some IXPs let all
  participants negotiate their own arrangements.
• This mesh network scales well, but it takes a
  lot more work by each ISP.
• If some participants choose not to peer with each
  other, then there will be a partial mesh instead
  of a full mesh.

20
Option 2 Bilateral peering

• Advantages
• Choose who to peer with or not
• All routers are managed by ISPs, not by IXP
  operator
• Complex policies are possible

• Disadvantages
• Non-peering may cause inefficient routing
• ISP router config becomes complex
• Difficult for new participant to connect

21
Option 3 Hybrid

• It is possible to have both models operating
  simultaneously at an IXP, with some ISPs peering
  with the central route server and the others
  manually configuring their routers for bilateral
  peering with selected peers.
• Not the most desirable!
• But can develop eg. if very large and very small
  ISPs are part of the same IXP gives control
  over business relationships.
• One option is to start with a single central
  route server and multi-lateral peering, and allow
  bi-lateral peering to be added later.

22
Exercise 1

• Several ISPs
• Each ISP has a router at HQ, linked to upstream
  provider
• Each ISP adds a new router at exchange point
  (XP), connected to HQ router
• Start using iBGP between HQ and XP
• Not yet peering with other ISPs

23
(No Transcript)
24
Exercise 2

• Follows on from exercise 1
• Each ISP starts BGP peering with all others
  (bi-lateral peering)

25
(No Transcript)
26
Exercise 3

• Follows on from exercise 1
• Undo part of exercise 2 first
• Each ISP starts BGP peering with a route server
  (multilateral peering)

27
(No Transcript)