BGP || IBGP

[Rajvendra pratap singh]

Hello All , 

As i learned in BGP that IBGP should be fully meshed , Means All router's in IBGP should have other's routers routes information.
As per split horizon rule : iBGP routers do not re-advertise routes learned via iBGP to other iBGP peers . 

So my confusion is here if a router learn a route from its IBGP peer and it will not re-advertise to other iBGP peers so how IBGP will become fully meshed in same AS . ? 

Please advise ?

[Ramblin Tech]

Sounds like confusion over what "fully meshed" means. In the iBGP sense, fully meshed means every iBGP speaker is peered with every other iBGP speaker in the same AS. That is, in each iBGP speaker there would be a neighbor statement for every other iBGP speaker in that AS. No need to re-advertise something learned from iBGP peer #1 to iBGP peer #2, as peer #2 would have learned the same prefixes from peer #1 in their own direct session.

In practice, iBGP peers are fully meshed in only a relatively small AS, as maintaining a large list of iBGP peers in each and every speaker does not scale. Instead, route reflectors are used so that an iBGP peering is only needed to the RRs (should be more than one for redundancy).

[MHM Cisco World]

R1-R2-R3

Full mesh meaning 

R1 have Neighbor R2 and R3 

R2 have neighbor R1 and R3 

R3 Have neighbor R2 and R1 

Note:-so R1 established neighbor with R3 through R2 ? Yes 

Note:- how R3 know R1 IP uisng for bgp neighbor? By config static or IGP

[Flavio Miranda]

Hi @Rajvendra pratap singh 

"So my confusion is here if a router learn a route from its IBGP peer and it will not re-advertise to other iBGP peers so how IBGP will become fully meshed in same AS . ? "

 Routers does not become fully meshed but redistributing routes but by using the command neghbord under BGP process. Which means, the fully meshed topology must be created by you and it will not be created automatically but routing table distribution.

[David Ruess]

Hello,

In the terms most are referring to as far as full-meshed and iBGP its mostly talking about eBGP learned routes (Border routers) or network/redistribution command learned routes. Think about it. Your BGP routers that peer with eBGP neighbors will learned eBGP routes. Then in turn those routes will be advertised to the iBGP neighbors since its was an eBGP learned route not iBGP (same also applied for network/redistribution command on the local router). Now those iBGP neighbors wont forward (by default) those networks to their iBGP neighbors. But if the border is fully meshed with every other iBGP neighbor in its domain then all the routers will have learned the route and have no need to forward it to their other iBGP neighbors since it'll already have that route. There are ways around this such as route-reflectors so you DONT have to be fully meshed. But as long as the iBGP 

Hope that helps

-David

[Joseph W. Doherty]

". . . so how IBGP will become fully meshed in same AS . ?"

That's for you to manually accomplish.

For instance, given the following hierarchical topology:

It would not be correct for iBGP unless the root iBGP (top) was a route reflector.

I added three links, to the above topology, creating a full mesh:

That's what iBGP needs to work, if you don't use router reflectors (or confederations).

The physical problem with full mesh topologies, the number of links scales exponentially, which is why (as noted by other posters), you don't normally do this unless you have very few iBGP nodes.

Basically, iBGP does not "relay" routes learned second hand, i.e. all other iBGP peers must be a connected peer.