BGP Convergence Time

aravindhs · ‎03-06-2008

Hi experts,

I have a question about BGP convergence time.

Consider an MPLS core and a customer on his own VPN (VRF) across the ISP's core network.

CE to PE is eBGP.

When I remove a route on the CE router, the other CE router (across the Core) detects it at once and removes it from the BGP table and the RIB.

But, when I add/publish a new route on the CE's BGP process, it takes a few seconds (say 30 to 50) for the other CE router to add to the BGP table and hence the RIB.

Is this common ? Does BGP take a while to publish routes to the other neighbor or is it because of delay across the core ? If it is, then why does the Withdrawals take place real quick ?

I did some debug bgp update events .. and I could see that the BGP UPDATES for new advertisements and withdraws happen instantaneously..

eg..

101043: Mar 6 13:25:35.480: BGP(0): 2.3.25.70 rcvd UPDATE w/ attr: nexthop 8.13.5.1/30, origin ?, path 5577 5577

101044: Mar 6 13:25:35.480: BGP(0): 2.3.25.70 rcvd 10.250.250.0/24

101045: Mar 6 13:25:35.484: BGP(0): 2.3.25.70 rcvd 8.13.5.1/30

101046: Mar 6 13:25:35.484: BGP(0): Revise route installing 1 of 1 routes for 10.250.250.0/24 -> 2.3.25.70 (main) to main IP table

101047: Mar 6 13:25:35.484: BGP(0): Revise route installing 1 of 1 routes for 8.13.5.1/30 -> 2.3.25.70 (main) to main

arav

Mohamed Sobair · ‎03-06-2008

Hi,

There is difference between Convergence and update..

Converge means when a router lost one or more of its paths to a particular destination.

If you are asking about BGP update, then yes , bgp update period takes 60 sec.

also consider that the route you have entered, should be propagated as an VPNv4 prefix through MP-BGP in the MPLS-backbone provider.

If you need changes to take affect immediately in BGP, you can reset the TCP connection between the bgp peers or configure (Soft reconfiguration).

HTH

Mohamed

aravindhs · ‎03-07-2008

Mohamed,

Thank you for your reply.

I could see that the default update timer is 30 seconds for eBGP and 5 secs for iBGP from the show ip bgp nei command output.

How are you saying it is 60 seconds ?

Also, I would say that routing convergence happens essentially because of the UPDATEs..

regards,

arav

Mohamed Sobair · ‎03-07-2008

Arav,

I mistakenly typed the keep alive timer, yes the update is 30 sec and keep alive is 60sec.

and a bout your point that you would say update is converge, Normally the fast convergence is associated with a routing protocol that detects a failur of a path very fast like eigrp,ospf.

offcourse, both would send periodic update when there is a new route added or removed, but with bgp this doesnt normaly happen, unless wit soft reconfiguration or peer reseting or takes the normal 30sec time for the new updates.

HTH

Mohamed

aravindhs · ‎03-07-2008

Hi Mohamed,

I just meant to say that the network convergence happens through 'UPDATE' messages (periodic, like RIP or triggerd/flash like OSPF,EIGRP).

But, with BGP there seems to be a deterministic delay associated with successive UPDATE messages.

I found this nice document when I googled.

cheers

Arav

Joseph W. Doherty · ‎03-07-2008

The eBGP default timers appear to have been tuned for Internet sized tables. When working within a MPLS VPN where routes now number perhaps hundreds, not hundred thousands, eBGP timers can be configured to much smaller values, sometimes all the way down to minimum allowance.