[c-nsp] iBGP confed question

[Pete Templin]

Tim Durack wrote:

> We run iBGP confeds between 3 POPs (seems more natural than
> route-relectors), redundant routers at every POP, topology is
> essentially a big ring over L2 transport.
> 
> iBGP sessions are between IGP loopbacks. (I always thought iBGP should
> be between loopbacks, rather than interface addresses, that way
> sessions survive link failures without effect.)

It's been 2-3 years since I abandoned confederations for route 
reflectors, and this is the first I've "looked back".  I can't remember 
specifically, but I think you might want to consider the ieBGP sessions 
(between the sub-ASen) on interface addresses, to allow the (sub) AS 
path to guide path selection in the event of WAN link failures.

> Is this correct, or is there an argument for iBGP between interface
> addresses, making BGP more "aware" of the underlying topology? Maybe
> it's a stupid question, but I haven't come up with a good answer
> yet...

FWIW, I switched from confederations to full mesh to route reflectors as 
we grew from four Riverstone routers to ~25 Cisco routers.  Our network 
was much happier as a result.  :)  Feel free to ping me offline if 
there's any guidance I can provide.

pt