[c-nsp] Varying CPU usage on 881

[Rodney Dunn]

That graph is total usage so there may be another process that
contributes that takes longer on the 881 that doesn't show up
on the 1841. ie: checkheaps, etc...

Watch 'sh proc cpu sorted' during one of those spikes and see
if you see another process contributing.

Also, I don't think it applies here given the pps number you gave
before. But I have seen in the labs where a test set clock syncs
up with the clock on the router and (without going in to too much detail)
the CPU shows busy when really it's not. It's how the CPU is measured
every 4 msec under the timer interrupt.

I usually see that when the rate is divisible by 4.

Rodney

On Mon, Sep 29, 2008 at 03:26:31PM +0200, Pelle wrote:
> Hi.
> 
> I'm doing some performance testing on a Cisco 881, but am a bit
> puzzled over a varying CPU usage. I don't see this behavior in for
> example a 1841. Have somebody got similar results? And (hopefully) an
> explanation of the variation? The variation in it self is no problem,
> but it makes it harder to do a performance evaluation.
> 
> In the tests, the same constant bidirectional bitstreams [1] are
> applied on both routers, and the routers are running the same
> configuration [2].
> 
> "show process cpu history" [3] on the 881 running 12.4(20)T [4]:
> 
>     333222221111133333222222222222222333332222222222444442222222
>     555999999999944444999991111188888777771111133333000001111155
> 100
>  90
>  80
>  70
>  60
>  50
>  40 ***                              *****          *****
>  30 ********     **********     **********          *****     **
>  20 ************************************************************
>  10 ************************************************************
>    0....5....1....1....2....2....3....3....4....4....5....5....6
>              0    5    0    5    0    5    0    5    0    5    0
>                CPU% per second (last 60 seconds)
> 
> Here the CPU load varies between 19 and 40 percent.
> 
> 
> The 1841 running 12.4(20)T:
> 
>     555555555555555555555555555555555555555555555555555555555555
>     444443333344444444444444444444444443333344444333333333333333
> 100
>  90
>  80
>  70
>  60
>  50 ************************************************************
>  40 ************************************************************
>  30 ************************************************************
>  20 ************************************************************
>  10 ************************************************************
>    0....5....1....1....2....2....3....3....4....4....5....5....6
>              0    5    0    5    0    5    0    5    0    5    0
>                CPU% per second (last 60 seconds)
> 
> The 1841 is rock stable with an CPU load of 53-54 percent [5].
> 
> 
> [1] Traffic consists of:
> - 8.0 Mbps @ 64 Byte/pkt downstream (14285 pps)
> - 8.8 Mbps in 3 classes upstream (2825 pps total)
> 
> [2] Configuration:
> - Downstream: just routing
> - Upstream: H-QoS with parent shaper (8.0M), 3 classes in child (1 LLQ)
> 
> [3] I know the "show process cpu history" shows both interrupt based
> load (traffic) and "other" load. Monitoring the SNMP variables
> cpmCPUTotalMonIntervalValue and cpmCPUInterruptMonIntervalValue shows
> that the "other" load never is more than 2-3 percent.
> 
> [4] The same results are achieved with 12.4(20)T1
> 
> [5] It's a bit surprising that the 881 outperforms the 1841, but
> nevertheless welcome :)
> 
> -- 
> Pelle
> _______________________________________________
> 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/