[c-nsp] BGP Cpu

[Jeff Fitzwater]

What platform are you running on?

We had similar CPU issue on sup-7203B which uses a TCAM for FIB entries. 
This is where the fast-switching occurs, and if there is not enough 
space then routes may be software switched more often causing increased  
CPU.

To see how the TCAM space is allocated, run the following...

*sho mls cef max*

The IPV4 entry is what you are most interested in.

Now check to see how many FIB entries there are...   You are most 
interested in the IPV4 entries.

*show mls cef summ*

If the IPV4 size is greater than or very close to the IPV4 size from the 
MAX  command then you are probable soft switching many packets and 
therefor high CPU.

We were right at the edge and were running 75% and now we are at 10%. 

The command to increase the size is ...

*mls cef maximum-routes ip* "new number goes here"  * NOTE: You must 
reboot for this to take effect*.

Ours was 192K and we raised it to 239K.

I might note that you are using other alloted TCAM space.  In our case 
we used some of the IPV6 space.

So with all that said, in you case when you do a "soft inbound" you are 
rebuilding the TCAM table, and the CPU probable does go down for a while 
since you are fast-switching most packets.

I hope this helped.


Jeff Fitzwater
OIT Network Systems
Princeton University

hjan at libero.it wrote:
>> 1) How many routes do you hold in your BGP RIB, from how many peers,
>>     
>
> 221060 network entries using 24979780 bytes of memory
> 884200 path entries using 42441600 bytes of memory
> 76839/47288 BGP path/bestpath attribute entries using 9220680 bytes of memory
> 1250 BGP rrinfo entries using 30000 bytes of memory
> 47832 BGP AS-PATH entries using 1259152 bytes of memory
> BGP using 78037472 total bytes of memory
> Dampening enabled. 0 history paths, 0 dampened paths
> 10 received paths for inbound soft reconfiguration
> BGP activity 20478750/20090098 prefixes, 109440806/108135132 paths, scan interval 60 secs
>
>   
>> etc?
>> 2) Are you using soft-reconfiguration? (I assume yes as you are
>> executing soft in).
>>     
>
> Yes, but mainly there are mpls vpn subscriber.
>
>   
>> Soft-reconfiguration maintains two sets of BGP routes on inbound receipt
>> from a peer. The first set is all routes sent from the peers, regardless
>> of any filtering. The second is the routes that pass the defined
>> filtering logic and are installed into the BGP RIB. 
>>     
>
> Ok this is clear, but what i can't understand is why cpu goes down after the clear and then in a couple of days it goes up again...
> I think this could be ok if you look at memory, but non on cpu side.
> However i try to tune some bgp timer and tcp path mtu discovery...
>
> Regards,
> Gianluca
>
>
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>