[c-nsp] BGP Tuning

[Phil Mayers]

On 04/01/2013 10:34 PM, Nick Hilliard wrote:
> On 01/04/2013 02:07, David Walters wrote:
>> Just a quick post to ask peoples experiences of tuning BGP convergence?
>> We're in the process of tuning the network to to achieve optimal
>> convergence (ie quick) and was wondering if anyones achieved this
>> outside of changing keepalive/holdtimers to 1 and 3. We're using 4900s
>> and playing with <100 prefixes.
>
> This doesn't sound sensible.  You should use a tightly tuned IGP (e.g.
> ospf) to handle your interior network connectivity and rapid convergence
> requirements.  BGP can then do the heavy lifting for prefix handling.

Although 1/3 second timers have always seemed a bad idea to me, slow 
detection of dead BGP sessions can still cause issues. OP should look 
into the availability of BFD or neighbour fall-over if they suffer from 
these.

You can simulate the latter with yucky EEM scripts and route track 
objects if your platform lacks "neighbor fall-over route-map" support 
(which most do...)