BGP Shut Vs Stop Advertisements

Unanswered Question
Sep 4th, 2011

Hi !

i have experienced in the past that when i take the bgp peer down [ shut ] , sessions are being affected .

when a bgp peer tears down, what happens to a session ?

i know in the bgp rib level it has another path to go to .

because there is only one best path, this means that this path only would be installed to rib-fib and the old one be removed.
and this i suppose - takes time .

my question is - during this time do the packets drop ? what happens to them ?

from the router's point of view, is there a difference between shutting the peer and stopping the advertisements ?

is that where and why cisco nsf/bgp graceful restart got into work ?

here is a scenario :

Cust peers to two different ISP's and advertise the same /24 prefx to both .

would like to move all in and out traffic to a single isp without interruption .

* no they cannot have a more specific prefixes to advertise to the other peer and thus "drying" one of the peers .

Thanks

I have this problem too.
0 votes
  • 1
  • 2
  • 3
  • 4
  • 5
Average Rating: 5 (1 ratings)
lgijssel Sun, 09/04/2011 - 06:40

This is an interesting question, one to which I did not know an immediate answer.

Fortunately, I had a lab (on GNS3) in which to verify this.

I placed a qemu host on the iBGP subnet between two routers providing the redundant WAN access.

DefGW was set to the primary router and the machine was made to ping to a loopback address on the WAN.

Then the BGP peer to the WAN link on the primary router was shutdown.

Contrary to my expectations, the ping went through undisturbed and without losing packets.

BGP timers in this lab are set to 10 30 but with other failover situations, I can clearly see an interruption of traffic.

Now, this is just a simulation and the behaviour may be different on a router hosting a full Internet table but the effect of shutting the peer seems to be that the routes are flushed immediately. My impression is that one cannot guarantee a total lack of downtime but shutting the neighbor via cli is obviously better than waiting for a protocol timeout.

Screendump of the test is attached.

regards,

Leo

Attachment: 

Actions

Login or Register to take actions

This Discussion

Posted September 4, 2011 at 3:18 AM
Stats:
Replies:1 Avg. Rating:5
Views:958 Votes:0
Shares:0

Related Content

Discussions Leaderboard