[c-nsp] Optimal solution for iBGP mesh

[Oliver Boehmer (oboehmer)]

Kanagaraj Krishna <> wrote on Saturday, April 14, 2007 8:35 PM:

> Hi,
>    As we all know, fully meshed iBGP topology is not suitable for
> network scalability. The recommended alternatives are route reflector
> and AS confederation (we are practising the former to a certain
> extent). At the moment we are thinking of having a few (maybe 2)
> routers just dedicated as Route Reflectors but the load on this
> routers would be high. 

How high would the load be? I.e. how many RR-clients, with how many
routes (full BGP table, I assume)? Which platform

> Just curious whether, MPLS can be used in the same way to reduce BGP
> sessions and processing load on network equipments?
> 
> - Maybe have internal routers using next-hop labels to direct traffic
> to edge routers.
> - Meanwhile only internal routers connected to downstream BGP
> customers are configured to receive full routes.
> 
> Please advise whether this option feasible. If there some of you
> guys/gals who are using a similar setup please share additional info
> on how to get it up and running (article, reference etc). Thanks.

This option is very feasible, and is used quite a lot. However, the main
reason for doing so is to reduce the number of routes on the core
routers (and thus speed up convergence, etc.), rather than to reduce the
load on the RR as the largest number of clients are usually the
PE/distribution routers, which would still need to receive the full
table, even with a BGP-free MPLS core.

Modern RR (like NPE-G1, fast ISRs, etc.) dedicated for this function can
easily sustain several hundreds of peers (if not more)..

	oli