[c-nsp] Peering between route reflectors

[Lee Clark]

For sure, minimum 2 sessions to 2 different RRs per client. I may have over-simplified the example to show why the IBGP session between RRs is needed. A more realistic example would have been 10 RRs, 100s of clients. In that case peering each client with all 10 RRs in the absence of an RR full mesh wouldn't be scalable.

Lee

-----Original Message-----
From: Mark Tinka [mailto:mark.tinka at seacom.mu] 
Sent: Monday, April 07, 2014 1:54 PM
To: cisco-nsp at puck.nether.net
Cc: Lee Clark; Cydon Satyr
Subject: Re: [c-nsp] Peering between route reflectors

On Monday, April 07, 2014 09:43:05 PM Lee Clark wrote:

> Without the RR/RR peering there is no way to propagate routes between 
> clients A and B. Peering both clients to both RRs that would solve the 
> problem but is not scalable in a large network where there are many 
> RRs and significant # of clients.

Agree that having 2x iBGP sessions per client scales poorer than one, but it scales better than a full mesh between routers, which is the problem route reflectors solve.

As you rightly point out, YMMV, but from where I'm standing, 2x iBGP sessions per client to 2x different route reflectors is fine for us. It's a reasonable compromise between redundancy and administration.

Mar.