[c-nsp] ibgp convergence time ?

[Jared Mauch]

On Sep 30, 2008, at 1:14 PM, julien leroiso wrote:

> Hi,
>
> I have 2 bgp routers (named border1 and border2) with a ebgp session  
> to one
> ISP (yyy.yyy.yyy.yyy) and ibgp session (xxx.xxx.xxx.1 and  
> xxx.xxx.xxx.2).
>
> My border2 is very very slow to converge. After +-3h I didn't get  
> all my
> border1 routes.
> I reset the session it's the same. I tryed to debug but I found  
> nothing
> interresting.
>
> Is it normal that ibgp is slower to converge than ebgp ?

	There's a lot of factors that can play into this.

	I had once seen a document that Cisco generated on ways to help with  
BGP convergence.

	Some of those methods are:

o Enable tcp path-mtu-discovery, to allow larger bgp messages.
o Increase the input hold queue from the default (70) to 4096.  This  
reduces the number of TCP retransmissions.

	I don't recall them all, but I can vouch for the above two.  I  
suspect you will see a marked improvement in convergence.

	It may be useful if you can identify the platform type (eg: 2600 vs  
GSR w/ PRP) and other things like system memory, etc..

	If you can provide more details, we can provide some better advice.

	- Jared