Why IGP in AS.

Unanswered Question
Sep 6th, 2010

Hi All,

Before asking question let me put here what I understood.

[ iBGP peers should not advertise routes learned from iBGP peers to other iBGP peers( as per BGP rules, which is there because AS-Path attribute functionality will be defeated..so this rule is good). The solution to this problem is  Full mesh among iBGP peers OR Route Reflectors OR Confederations.]

We know lot of ISP's run IGP inside their core even though they have iBGP running. Why do they need this IGP(ospf or isis) if they have satified above required for iBGP.

My question is under what conditions IGP must be used while iBGP is serving well enough ?


I have this problem too.
0 votes
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Jon Marshall Mon, 09/06/2010 - 06:12


You are correct when you say that if you have a full mesh IBGP you do not need an IGP within the AS. However this assumes you are using the physical interface addresses to peer with IBGP and best practice is to use loopbacks for the peering. If you use loopbacks you do now need an IGP so that each IBGP router knows how to reach the other routers loopbacks.


Chetan Kumar Ress Mon, 09/06/2010 - 06:28

Hi Ganesh

Adding to Jon

1] Yes to advertise Loopback interface you need an IGP to form IBGP neighbor using loopback which give you more Flexibilité.

2] In ISP network the other reason is to have IGP becasue of presence P router, P give you more flexibilite to update all PE Routers ( If P  is configured with RR) that connect to CE router becasue the BGP update use loopback & That is reachable via IGP.

3] If service provider run without RR then you should have Full mesh as you told. BUt for full mesh you need that much of link which increace you capex, In that senario you need an central router or location know an P router or HUB  where all link are terminated & every router configured with Loopback IP to form an full mesh BGP & P router won't need to configure BGP it run only IGP i.e. OSPF  or ISIS.


Chetan Kumar


Giuseppe Larosa Mon, 09/06/2010 - 07:41

Hello Ganesh,

the IGP is almost needed in all SP networks:

advertising the loopback IP addresses is a pre-requisite for signalling plane:

iBGP uses it

LDP uses it

RSVP TE uses it

L2TPv3 uses it

and so on

When MPLS is used you need a working IPv4 cloud to exchange label information with LDP or RSVP TE.

This is a divide and conquer approach:

otherwise an iBGP session state would depend on the state of other iBGP sessions and this is not desirable from a troubleshooting point and for more robustness.

OSPF or IS-IS are very good in doing the job of advertising loopbacks and backbone links and they do it efficiently with fast convergence.

iBGP sessions can be seen like user traffic flows that use routing table entries built by IGP and eventually MPLS paths.

Actually in an MPLS cloud iBGP packets can travel within an MPLS LSP with no problem like user traffic.

I have seen this in packet captures.

Hope to help


james-worley Tue, 09/07/2010 - 02:57

What he said!

The IGP should carry your infrastructure routes and loopbacks, be kept as small as possible and thus be quick to converge. BGP builds upon and relies upon your IGP for next-hop reachablility - this becomes self evident in a multipath IGP scenarios where more that one route exists between two iBGP routers.

BGP is slower to converge and carries customer routes.

The old Cisco ISP Essentials book is a great read on this subject and is available as a PDF with a few good google searches.

;-) James


This Discussion