Slow BGP convergence

[j0nathanj]

Hi,

I set up Catalyst6500 Quad-SUP VSS system.
I have a problem with slow BGP convergence(80sec) on VSS switchover.
Each chassis has a bgp peer and "bgp graceful-restart" is disabled.
I investigated BGP behavior with "debug ip bgp events" and "debug ip bgp update".
I found that BGP process waited a peer to reply until timeout and after the timeout BGP table was calculated.
I guess this timer makes the convergence slow.
How can I adjust this timer value?

----------------- debug message ------------------
*Aug  8 14:56:09.989 JST: BGP(0): 1.1.1.1 was the first peer to be established for IPv4 Unicast
*Aug  8 14:56:09.989 JST: BGP: nopeerup-delay post-boot, set to default, 60s
*Aug  8 14:56:09.989 JST: %BGP-5-ADJCHANGE: neighbor 1.1.1.1 Up
*Aug  8 14:56:09.989 JST: BGP(IPv4 Unicast): waiting on established peer 1.1.1.1 EOR
*Aug  8 14:56:09.993 JST: BGP(0): 1.1.1.1 rcvd UPDATE w/ attr: nexthop 1.1.1.1, origin i, metric 100, merged path 11111, AS_PATH 11111
*Aug  8 14:56:09.993 JST: BGP(0): 1.1.1.1 rcvd 0.0.0.0/0
*Aug  8 14:56:10.509 JST: BGP(IPv4 Unicast): waiting on non-established peer 2.2.2.2 for 70s (initial peer up)
*Aug  8 14:56:10.785 JST: BGP(IPv4 Unicast): waiting on non-established peer 2.2.2.2 for 70s (initial peer up)
*Aug  8 14:56:11.533 JST: BGP(IPv4 Unicast): waiting on non-established peer 2.2.2.2 for 70s (initial peer up)
*Aug  8 14:56:12.557 JST: BGP(IPv4 Unicast): waiting on non-established peer 2.2.2.2 for 70s (initial peer up)
*Aug  8 14:56:13.581 JST: BGP(IPv4 Unicast): waiting on non-established peer 2.2.2.2 for 70s (initial peer up)
- snip -
*Aug  8 14:57:19.117 JST: BGP(IPv4 Unicast): waiting on non-established peer 2.2.2.2 for 70s (initial peer up)
*Aug  8 14:57:20.141 JST: BGP(IPv4 Unicast): computed bestpaths, table version went from 10 to 34
----------------- debug message end ------------------

[Alessio Andreoli]

Hi Jun,

please find these references:

http://ieoc.com/forums/p/6115/21381.aspx

and

http://stack.nil.com/ipcorner/DesigningBGPNetworks/

but keep in mind that BGP has been designed to be slow and it is NOT a good protocol to replace the job of an IGP.

Hope it helps

Alessio

[j0nathanj]

Hi Alessio,

thank you for your reply.

I didn't write that all BGP peers are directly connected EBGP neighbors.

And I've checked that "fast-external-fallover" is enabled by default.

In this case, a BGP initialization occurs on the routing engine(supervisor) switchover.

Why does the BGP process wait the peer connected to rebooting chassis in the BGP initialization process?

Is there any workaround?

[John Blakley]

You can adjust your timers globally or per neighbor:

router bgp 1

timers bgp 7 21

OR

router bgp 1

neighbor x.x.x.x timers 7 21

I wouldn't recommend anything lower than 7 for a keepalive if this is an ebgp peer. For ibgp, I set mine to 5 and 15. The lower of the of the 2 timers is used between the peer, but the peer needs to be cleared before the new timer goes into effect.

HTH,

John

[j0nathanj]

Hi John,

thank you for your reply.

I set BGP timers to "10 30", but this didn't solve the problem.

[j0nathanj]

Hi,

I solved this problem by setting "bgp update-delay xx".

[bonger0330]

What bgp update-delay value did you set to improve convergence time?