[j-nsp] IBGP problem

Pedro Roque Marques roque at juniper.net
Wed Oct 12 06:45:59 EDT 2005


Nils Kolstein wrote:
> Hi,
> 
> We encounter a strange problem regarding an IBGP ession between two
> M160's. Both run JunOS7.3R1.6. The IBGP session between the two stays up
> only for about 1,5 minutes after which it drops again. What I notice is
> an increase in the OutQ-value:
> 
> xxx.xxx.xxx.xxx    xxxx          2        174   35413       4
> 1:33 Establ
>                                               ^^^^^
> The log says:
> 
> Oct 11 08:50:01 bgp_traffic_timeout: NOTIFICATION sent to
> xxx.xxx.xxx.xxx (Internal AS xxxx): code 4 (Hold Timer Expired Error),
> Reason: holdtime expired for xxx.xxx.xxx.xxx (Internal AS xxxx), socket
> buffer sndcc: 16367 rcvcc: 0 TCP state: 4, snd_una: 746157955 snd_nxt:
> 746162051 snd_wnd: 57192 rcv_nxt: 2416270658 rcv_adv: 2416327867,
> keepalive timer 0
> 
> This problem seemed to have popped quite sudden without any changes made
> in BGP setup or otherwise.
> 
> Does anyone has a clue?

You most likely have a connectivity problem w/ large sized packets on 
that path. i.e. small packets w/ Hellos can make it thought but not UPDATEs.

ping tos 0xc0 size <x> remote-host will probably let you figure out the 
problem.

Check your mtu-discovery setting also... if it is enabled and mtu 
discovery is not working figure out what packet sizes TCP is using by 
using tcpdump on the RE-orinated traffic.

   Pedro.


More information about the juniper-nsp mailing list