[c-nsp] Non-default BGP hold / keepalive timers

[David J. Hughes]

I'm just looking for other peoples experiences with reducing the hold 
down time of BGP peers to improve convergence time when a peer just 
disappears on you.  As ethernet delivered services are becoming more 
popular the chance of having end-to-end link with the actual peer 
device is getting smaller.  If the peer dies, blackholing traffic for 
the default 180 seconds just isn't good enough.

What are people running on their eBGP borders?  I've read of some very 
aggressive timer use for iBGP sessions (i.e. running 3 * 1 second 
keepalives) but I wouldn't run that internally, let alone with an 
external peer.  I can't find any BCP or similar that covers this.  I 
don't really want to start tweaking them lower an lower until things 
start to flap.

I am thinking about 3 * 5 seconds giving a 15 second outage in this 
situation.  Thoughts?


David
...

cisco-nsp at puck.nether.net