Re: rt for newbie

[Carsten Emde <Carsten.Emde@osadl.org>]

Sashka,

> Did I understand you correctly, I need to disable Dynamic Ticks and
> Check for stack overflow?
Yes.

> If yes, then here are my results
Thank you.

> # cyclictest -m -Sp99 -i100 -d0
> # /dev/cpu_dma_latency set to 0us
> policy: fifo: loadavg: 0.01 0.09 0.05 2/122 2187
>
> T: 0 ( 2180) P:99 I:100 C:1969005 Min:      1 Act:    2 Avg:    1 Max:      10
> T: 1 ( 2181) P:99 I:100 C:1968946 Min:      1 Act:    2 Avg:    2 Max:      69
> T: 2 ( 2182) P:99 I:100 C:1968888 Min:      1 Act:    2 Avg:    2 Max:      12
> T: 3 ( 2183) P:99 I:100 C:1968829 Min:      1 Act:    2 Avg:    2 Max:      52
> T: 4 ( 2184) P:99 I:100 C:1968770 Min:      1 Act:    2 Avg:    1 Max:      10
> T: 5 ( 2185) P:99 I:100 C:1968712 Min:      1 Act:    5 Avg:    2 Max:       9
> T: 6 ( 2186) P:99 I:100 C:1968654 Min:      1 Act:    2 Avg:    2 Max:      55
> T: 7 ( 2187) P:99 I:100 C:1968594 Min:      1 Act:    2 Avg:    1 Max:      17
>
> CPU I've got is Xeon X5472  @ 3.00GHz
These data look very promising.

In a next step, you may want run a billion cycles (-l1000000000) which 
takes a bit more than a day and apply a load scenario that reflects the 
load to be expected under production conditions. This will provide 
already a good estimate of the worst-case latency of the system.

The worst-case latency can be reduced further, if you disable throttling:
# cd /sys/devices/system/cpu
# for i in cpu*/cpufreq/scaling_governor
 > do
 > echo performance >$i
 > done

	-Carsten.