[j-nsp] BGP problem...

Michael Loftis mloftis at wgops.com
Thu Apr 7 12:15:05 EDT 2005



--On Thursday, April 07, 2005 2:31 PM +0300 Erol Kahraman 
<erol.kahraman at gmail.com> wrote:

> hi,
> I have a M7i router which is connected to internet via 2Mb F/R
> connection. Everything is ok. But when i make a stress test my BGP
> connection peer send me log as follows.
>
> Apr  7 12:48:20  Router1 rpd[2520]: bgp_traffic_timeout: NOTIFICATION
> sent to x.y.z.t (External AS 21): code 4 (Hold Timer Expired Error),
> Reason: holdtime expired for x.y.z.t (External AS 21), socket buffer
> sndcc: 19 rcvcc: 0 TCP state: 4, snd_una: 3435220949 snd_nxt:
> 3435220968 snd_wnd: 16384 rcv_nxt: 1777406639 rcv_adv: 1777423023,
> keepalive timer 0
> Apr  7 12:49:02  Router1 rpd[2520]: x.y.z.t (External AS 21): reseting
> pending active connection
>
> I have flooded an UDP packets to any server in my network from
> external source. I have filter which block that traffic but after some
> time the BGP connection is reseted. After that i've wrote a policer
> like this to limit all traffik to 1,9M and exclude BGP packets for the
> rest 100k bandwith.
> In spide of this my problem is persist. Do some of you have
> experienced problem like this.
>

>                     log;
>                     reject;

Don't reject, discard, and don't log either. reject sends packets BACK 
flooding the other side of your link too (unless this is what you want to 
do for the test) and log can overwhelm the RE in a situation like this. 
And by Mb i'm not sure if you mean Mbyte or Mbit either...at Mbit i find it 
hard to believe you'd be overloadign anythign other htan the FR circuit. 
at Mbyte it's possible the RE is getting overloaded.  Also since the filter 
and policer is on the input on your side, nothing stops the other side from 
fillign the pipe too full with UDP packets so that the BGP session goes.




More information about the juniper-nsp mailing list