[c-nsp] BGP routes co-existing with different local-preference

[David J. Hughes]

On 25/10/2005, at 9:29 PM, Kristofer Sigurdsson wrote:
> Upon further investigation, it seems that the BGP session with the
> router that went down took it's time to time out, which meant the
> former best routes stayed for some time.

Yup, we had a thread on this recently.  By default an IOS based box 
will require between 3 and 4 minutes to detect that the other "not 
directly attached" router has gone away (i.e. no interface state 
change).  This is the result of IOS using 60 second keepalives for the 
BGP session and you need to miss 3 keepalives before the hold timer 
expires.  Then factor in time for other paths to be advertised and 
selected as best and you can be down for quite some time.

We are running trials with reduced keepalive times and things are 
looking good.  I'll post a summary here once I have more confidence in 
the results.  This is something you don't want to just randomly mess 
with.


David
...