[c-nsp] Peering between route reflectors

[Peter Rathlev]

On Mon, 2014-04-07 at 21:01 +0200, Cydon Satyr wrote:
> Basically, if there are multiple route reflectors NOT in the
> forwarding path of the traffic, is there ANY reason to peer between
> them? I don't see a reason why they should peer, but I'd like to get
> this confirmed.

We (enterprise setup, ~10k routes and very little churn) peer between
ours, simply because I was always taught that iBGP non-RR-clients need
full mesh. Now that you ask I can't put my finger on exactly why they
should though. (Apart from RFC 4456 saying so of course.)
 
> Also, if they are NOT in the forwarding path, regardless of whether
> they are peering between themselves or not, it shouldn't matter if
> they are all in the same CLUSTER, correct ?

You should not make them one cluster, since it introduces black-holing
in certain failure scenarios. Let all clients peer with two RRs in each
their cluster.

A quick Google-search presents an explanation:

http://rekrowten.wordpress.com/2013/05/24/bgp-route-reflector-and-why-is-cluster-id-obsolete/

-- 
Peter