[nsp] Monitoring BGP links

[Steve Francis]

So last night we had an "interesting" failure. One of our peerings 
transits a fastethernet metro- area provider.  This virtual circuit 
degraded in such a way that most traffic failed to transit it - but 
enough made it that the BGP peering stayed up, causing the peer on the 
far end to keep advertising our routes, but then pumping  the traffic 
onto the circuit  where most of it got lost.

How do others guard against this?
- Use only "normal" circuits (OC3, OC12, or direct long haul ethernet) 
that tend to degrade in a more binary fashion (other than congestion). 
Now it works, now it doesn't.
- Use something like response time monitor to measure performance and 
packet loss directly from the router to the router on the other end of 
the link?
- Creative tweaking of BGP keepalive timers, or circuit keepalives?

Open to ideas here.
Thanks