[c-nsp] BGP - Multihop across igp network

[Paul Stewart]

Thank you Sascha and Jon for your responses.... makes much more sense now...

We have the capability to bring up iBGP on the 6500's which we might examine
(just never needed to).  I think it's safe to assume in our environment that
if we brought up full tables on the 6500's using iBGP that there is no need
to redistribute anything into OSPF as they will become aware of the route
anyways.

If we didn't do that, would it be best to redistribute on the 7600's (EBGP
to OSPF) or would it be better at the 3825 side of things (which we manage
and have full control over - it participates in OSPF as well)?  I'm thinking
it would make more sense at the 3825 side?

Cheers, and thanks again for making this more logical for me..;)

Paul


-----Original Message-----
From: Sascha E. Pollok [mailto:nsp-list at pollok.net] 
Sent: Tuesday, April 06, 2010 11:03 AM
To: Paul Stewart
Cc: cisco-nsp at puck.nether.net
Subject: Re: [c-nsp] BGP - Multihop across igp network

Hello Paul,

> Connected to these 7600's we have a pair of 6500's not doing BGP (however
> participating in OSPF as all our boxes do)
>
> Connected off one of the 6500's we have a 3825 at a customer site
>
> The customer who is connected to a 3825 wants a full BGP feed from us.
>
> So, I created a pair of ebgp-multihop sessions between the 3825 and our
two
> 7600's - both tables populated and everything looked good.  Once the
> customer brought up their session all the traffic seems to loop between
our
> networks.. Almost like a blackhole effect.  In our 7600's I can see the
> customer routes advertising out and BGP itself looks correct when I do a
> "show ip bgp xxxxx" towards one of their netblocks.

Your other routers need to know about the customer's network.
You can either redistribute the customer's EBGP prefixes
into OSPF (selectively using e.g. route-maps on the 7600's)
or talk BGP to your other routes but learn only the customer's
routes. We are doing a similar thing. We are tagging customer-learned
routes with a BGP community and announce only prefixes carrying
this community via IBGP to the access routes.

Thus, the access routers carry only a few prefixes and can perfectly
route towards the customer.

Sascha