cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
947
Views
5
Helpful
7
Replies

Cisco 7604 normal cpu usage?

antp
Level 1
Level 1

Hi!

Well, we have cisco 7604 with rsp720-3c-ge running ios Version 12.2(33)SRB1, RELEASE SOFTWARE (fc3) and ws-x6724-sfp+dfc3c, 4 upstreams with bgp full-view and about 20 peers to announce full view.

This device routes about 1Gbit/s @ 0,2 mpps.

And CPU usage is 50-70%.

"BGP Router" process just about constantly running and using 10-15% of CPU.

dCEF enabled and running without exception.

Is that state normal? Will rsp720-3cxl work better (have lower cpu utilization)?

How we can improve perfomance for switching of 5Gb/s without high load and delays?

7 Replies 7

Joseph W. Doherty
Hall of Fame
Hall of Fame

If not already using it, you might want to investigate usage of BGP peer-groups to reduce impact of working with so many BGP peers. See http://www.cisco.com/en/US/tech/tk365/technologies_tech_note09186a0080093fb7.shtml

antp
Level 1
Level 1

Ok, may be we can reduce usage of CPU by "BGP Router".

But what to do with other 40%?

If we will exchange our rsp720-3c-ge to rsp720-3cxl-ge would it help in future?

From the specs of the xl, it runs at the same speed as the non-xl, so I wouldn't expect it to reduce the CPU, but with the additional resources on the xl for dealing with many routes, not 100% sure it wouldn't help.

I also wonder whether different line cards would make any difference, such as the SIP series using Ethernet ports. It might depend on what is causing high CPU and what the line card's local processor supports. Again, not 100% sure, but ordinary DFCs might only handle DCEF where some of the SIP cards might be able to off-load more. (I know some of them support QoS features otherwise unavailable.)

Lastly, also not 100% sure the group BGP peering reduction only would be against "BGP router". From Cisco's document I referenced in my prior post, "Based on the number of peer group members, the number of prefixes in the table, and the number of prefixes advertised, this can significantly reduce the load." Since this is just a configuration change, if this feature is supported on the RSP, something you could try before swapping supervisor or line cards.

Thanks for advance! Would try to group peers.

BGP peer-grouping slightly helped. Thanks.

After investigating we found that main CPU load was from using policy-routing.

If we removing it - load is about 10%.

Have any suggestions why it`s so boring for CPU? Shouldn`t policy-routing work at PFC level?

Thanks again!

Again, not 100% sure, but I recall reading some PBR statements have a higher impact than others, perhaps your policy had statements that caused some form of process switching. If you haven't already, and if supported, you might also try policy cache and/or flow cache on your interfaces.

kimpin
Level 1
Level 1

HI, could I check since you are using RSP720-3C for full internet routing, do you hit any problem with users not able to reach certain prefixes since non-xl could only takes 256000 routes? Thanks.

Getting Started

Find answers to your questions by entering keywords or phrases in the Search bar above. New here? Use these resources to familiarize yourself with the community: