BGP NSF/Graceful Restart

vinbuck · February 2012

Hey guys...wanted to throw out a question for those of you who work with MP-BGP on 7600 or similar chassis.

I'm working on enabling BGP graceful-restart throughout the network (OSPF, iBGP/MPLS with RR and VRFs), but wanted to get an idea of the impact on the control plane/forwarding plane before I enable it (will most likely be in a maintenance window). I don't have a spare 7600 laying around to test this on and GN3 ain't gonna cut it.

I've been going through Cisco docs to try and get an idea and they all speak to the theory of it but not the actual impact when you implement it. Also, saw mention of enabling 'ip cef distributed" as a prerequisite but not all of the routers i'll be enabling it on will be running cards that have distributed CEF cpability...any thoughts on that?

I've already got nsf running on OSPF, so hopefully enabling graceful restart will take full advantage of the NSF capability of our RSPs....at least that's the plan anyway

wirerat · February 2012

I am pretty sure I had read somewhere in the NSF docs that you will have to reform the adjacencies after implementing NSF because it is negotiated at the beginning of the peering. Let me see if I can find what I read.

Edit:

If you enter this command after the BGP session has been established, you must restart the session for the capability to be exchanged with the BGP neighbor.

Pg. 14
http://www.cisco.com/en/US/docs/switches/lan/catalyst6500/ios/12.2SX/configuration/guide/nsfsso.pdf

vinbuck · February 2012

Thanks for the reply...I did a bunch of reading yesterday and discovered that MPLS has its own graceful restart protocol that must be enabled for everything to play well together when SSO occurs. For those of you who may need to implement this on a similar provider network, you'll need:

1) ip cef distributed enabled (it will revert to centralized operation if you don't have distributed capable cards)
2) 'nsf' - command under your OPSF process (or whatever IGP you're using)
3) neighbor {neighbor peer group or IP} ha-mode graceful-restart - command under router bgp AS#
4) mpls ldp graceful-restart - command in global config

The BGP portion can be enabled under the entire routing process or by neighbor/peer-group. I chose to do it by peer group just to have a little more control should it be necessary in the future. And as wirerat mentioned, you will need to refresh your BGP sessions. Luckily, I was enabling BGP authentication at the same time, so I got a two for one deal as the peers dropped off and authenticated.

vinbuck · February 2012

So after doing some testing, I got mixed results. I forced a switchover on my test router and it did preserve the adjacencies for OSPF, BGP and MPLS and continued forwarding packets. However, after waiting about 15 minutes for all the aftershocks of the switchover to settle down, I tried to restore my router to the RSP it was previously using and it dropped about 4 packets. After examining the neighbor relationships on BGP, it appears that it wasn't able to preserve any of the address families in BGP during the switchover.

 Neighbor capabilities:
    Route refresh: advertised and received(new)
    Address family IPv4 Unicast: advertised and received
    Address family VPNv4 Unicast: advertised and received
    Graceful Restart Capability: advertised and received
      Remote Restart timer is 120 seconds
      Address families advertised by peer:
 [B]       IPv4 Unicast (was not preserved), VPNv4 Unicast (was not preserved)[/B]

I've been digging through the router and the web to figure out why, but so far no luck. Any of you guys have some suggestions of where to look to begin figuring this out?

vinbuck · June 2012

Hate to dredge this one up, but anyone have a thought on it....still researching and haven't found much about this output of sh ip bgp nieghbors

[B]IPv4 Unicast (was not preserved), VPNv4 Unicast (was not preserved)[/B]

BGP NSF/Graceful Restart

Comments