[c-nsp] Redistribution BGP/OSPF best practice

[Nathan]

On Mon, Feb 18, 2008 at 7:52 PM, Nathan <have.an.email at gmail.com> wrote:
>  For now I'll change OSPF costs so nothing goes through there (not
>  trivial unfortunately since it's a very central link) and program a
>  reboot of B.

Rebooting B fixed the problem, the problem being that A and B did not peer LDP.

B is a relatively old router (end-of-lifed), has an IOS without (I
think it's called) "clear mpls ldp",  and had recently seen some
overloads through too much traffic, so I think the incident is closed.

>  > > and OSPF? I redistribute some internal routes between BGP and OSPF,
>  > > but the why and the "how to avoid" of that is a story in itself.

This problem is coming back to bite me again. I have routers in
different datacenters, all interconnected by iBGP, and clients
connected to different datacenters using my IPs, with private AS eBGP
for redundancy. When I privilege one link, I set either
local-preference on my side, or (preferably) MED on the client side.
When I do not privilege one link, I want the backbone OSPF cost to
determine what link is used. Using BGP only it doesn't happen that
way, because BGP does not know what exit router is nearest. So I
redistributed all my IPs from BGP to OSPF (I don't have enough IPs for
that to be a volume problem is OSPF). Then it seems to work, but only
as long as the link hasn't flapped, since when the primary link goes
down, the secodary begins to redistribute into OSPF, and when the
primary comes back up it does not install the routes because OSPF
routes already exist.

What is the best practice solution to this problem? Preferably one
that lets me stop redistributing BGP into OSPF; am I correct in
thinking that that was a bad idea to begin with, and that there is no
good reason to do it?

Thanks,
--
Nathan