BGP performance tuning

[syannetwork]

Hi,

Currently have a BGP multihoming configuration to 2 different ISPs. BGP failover time seems to be taking between 3/4 mins. All BGP timers are currently set to defaults. Can anyone give me any advice on how to reduce the time it takes for BGP to failover onto the secondary eBGP peering when the primary eBGP peering stops forwarding traffic? I know we could possibly change the keepalive and holdtimes on the eBGP neighbors, but not sure this would be good practice.

Any recommendations or advice would be greatly appreciated.

Thanks!

[Giuseppe Larosa]

Hello Simon,

ebgp external fallover can help to sense if the link to eBGP neighbor fails.

and newer neigh fall-over using BFD

http://www.cisco.com/en/US/docs/ios/iproute/command/reference/irp_bgp3.html#wp1013842

reducing the keepalive interval if the number of peers is low has low impact.

Hope to help

Giuseppe

[syannetwork]

Giuseppe/Joseph,

Many thanks for the advice! I have tested 'neigh fall-over' and reduced the BGP timers on our lab kit and this has dramatically reduced the time it takes for the eBGP peer to age out, and BGP to converge over to the secondary link.

I will now be discussing this with our ISPs who we have eBGP peering arrangement, to look at the possibilities of implementing this solution on the live service.

Once again, thanks for the help guys!

[Joseph W. Doherty]

Personally, with current equipment and links, multiple BGP paths and interactive to real-time traffic, I've felt default BGP "neighbor timers" of 60/180 sec is slugish. This is something you might discuss with your ISPs; determine their thinking about "best practice" regarding reducing the values.

Besides what Giuseppe suggests, another newer technology that you might want to consider is OER/PfR, if your platforms support it. This can often fairly rapidly detect problems to specific prefixes and route traffic (best outbound) to another path. (The technology can do much more.)

[huangedmc]

Luckily, even if your ISP doesn't want you to fine tune the BGP timers, there's nothing to stop you from doing it.

BGP will use whatever's the smaller value.

This means if ISP insists on 60/180, but you decide to do 10/30, your BGP session will do 10/30.