[c-nsp] BGP next-hop, default routes and more!

[Jee Kay]

I am currently trying to set up the following routing for our CPE bits:

CPE--PE--Aggregation--Core

The Core device generates our aggregate routes and does BGP RR magic.
Our Aggregation connects to said RR magic, and the CPE then connects
to the Aggregation layer via IBGP to learn various routes. In
addition, the CPE has a static default configured pointing at the
remote PE interface.

So far so good..

The problem is that that the aggregation device loopback is in the
address space covered by the aggregate routes advertised by the core.
As far as I can tell, this leads to:

a) CPE uses default route to set up IBGP session with Aggr. so far so good.
b) CPE sees IBGP route which is 'better' (than default) for our
aggregate address space. Protocol next-hop is the Core device. Actual
next hop is obviously the PE.
c) Best route to Aggregation device suddenly becomes the IBGP learnt route
d) Endless cycling of routes as BGP routes alternately become
cyclic/inaccessible, are removed, are re-learnt, become cyclic, are
removed, etc.

at this point it all goes a bit wrong and the routing table basically
ends up in constant churn.

The question is.... how do I avoid this? The only real way I can think
of is to configure static /32 routes for the aggregate RRs and the
core RRs so that there is always a next-hop for both the BGP peering
and for any core generated routes. This sort of sucks though as it
would involve reconfiguring all CPE should the core addresses ever
change.

I assume I'm not the first one to run into this so... help? :)

Thanks,
Ras