From: Joseph L. Brunner (joe@affirmedsystems.com)
Date: Wed Jan 28 2009 - 08:14:05 ARST
The question was: "Why do we need a full-mesh between _all_ the
BGP-running routers in an AS?"
>marc, In an AS would assume ibgp, no? sure confederations are technically in an "AS" but they have rules to preserve some ibgp like behavior, med, others come to mind...
(we're not talking about confederations
or RRs here - just basic BGP).
>yup
He proposed the following alternative:
>I love when students are eager enough to do this (can you spare some great students like this!?)
-we have an AS with edge routers (running both iBGP and eBGP), and
transit routers (iBGP only).
-we do full-mesh BGP between the edge routers, and only connect each
transit router to each edge router.
>hmmm this sounds a lot like route reflectors... actually I teach rr this way...
-for 4 edge and 8 transit routers, this takes the number of BGP
connections from 12*11/2=66 to 4*3/2+8*4=38. Quite a significant change.
>we actually design ibgp with rr's quite a lot this way... the rr's have only their local peers and their other route reflectors as peers... so edge ebgp/ibgp routers at the pops only peer with the "backbone" which are route reflectors... google "hierarchical route reflectors" should shed some light on what I'm talking about on a large scale.
Do this, make a lab to show them the use of a cluster-id and it will become apparent- all this was thought of and hashed out a long time ago ;)
-Joe
Blogs and organic groups at http://www.ccie.net
This archive was generated by hypermail 2.1.4 : Sun Mar 01 2009 - 09:43:40 ARST