Cisco BGP Scanner CPU -Help

[bbisp]

I currently have a Cisco 7206VXR (256M RAM - IOS 12.2(3) enterprise) running BGP over ATM STM-1 and FA0/0 utilising 802.1Q. Currently it has 10 peers configured, of which 3 are IBGP (approx 10,000 routes), 1 eBGP (105,000 routes), 3 partial peers (20,000 routes) and 3 Customer peers (100 routes). When viewing BGP CPU utilisation (show proc cpu | include BGP), the BGP scanner is constantly at 50-70% utilisation. The router becomes CPU bound every 10-15min causing all OSPF and BGP sessions to restart. During this period all routing is lost.

When removing the 1 upstream peer (105,000 routes), the router stabilizes and the BGP scanner CPU usage returns the 0-3%. Can anyone suggest what could be the cause? Cisco Engineers, maybe?

p.s. I have soft-reconfiguration inbound configured on all BGP connections. I have 6 OSPF sessions configured with a total of 500 routes. No protocol re-distribution between BGP and OSPF. Static routes are re-distributed into OSPF.

[ben.l]

Which NPE/NSE and I/O Controller in your 7206VXR's, this should be more effect on the performance.

[bbisp]

Gent,

Thanks for your response. I have the NSE-1 R7000 CPU type. I think you might of hit the cause on the head. I am investigating this avenue. Do you know of any past issue with NSE-1 controllers and high CPU load when handling the Internet routing table?

[mksmith]

Hello:

I would check a couple of things.

1) Make sure "no synchronization" is listed under your BGP router process

2) Disable Soft Reconfiguration

Also, not to be picky, but you are kind of on the bleeding edge running 12.2 code on a production box. You might want to consider backing off into something in the 12.0.xS train or the 12.1.xT train (whatever is appropriate for your hardware configuration).

Mike

[bbisp]

Mike,

Thanks for your response. No synchronization is enabled under the BGP config. Disabling soft-reconfiguration will cause me a lot of heart-ache. Have you seen a problem like I have, fixed by disabling soft-reconfig?

IOS 12.2(3), has been tested with my configuration thoroughly and it has proven very reliable. I am running this version for the last 6-8 months. Any thoughts?

[ben.l]

The IOS version is more effective, if you can get the IOS release 12.0(18)S3 image, it's great, because, some big ISP using this image, and this is recommended by TAC to ISP, so, I belive it's a stable release(Note: Do NOT use the 12.0(19)S images).

Another recommendis is separate the BGP and Interdomain routing from one system.

BTW, could you please execute some show command and past the result on this board:

show ip bgp summary

show ip bgp neighbor xxx.xxx.xxx.xxx

(where the xxx.xxx.xxx.xxx is the ip address of you service provider's BGP peer.)

[bbisp]

Ben,

Thanks for the tip re: IOS version.. will investigate.

In regards to posting the results of the above commands, I am happy to do so (with some changes obviously). I am not sure how the formatting will turn out, OR if everyone else is interested.

So, if I get a number of interested people, I will post it, otherwise I will be happy to respond personally.

Could the people on the board indicate their interest.??

[6mmarinzulich]

Problem is related with your full routing peer. So I suggest you enable BGP dampening. Also to prevent BGP peers flapping disable fast fallover. Include these commands in your BGP configuration:

no bgp fast-external-fallover

bgp dampening

Good luck

Matias.-