Re: [patch] Real-Time Preemption, -RT-2.6.9-mm1-V0

[K.R. Foley]

Ingo Molnar wrote:
> * Ingo Molnar <mingo@xxxxxxx> wrote:
>
>
> > > Still problems with interactive behavior. Running KDE, with top
> > > running in xterm, scrolling through the menus I get some pauses. When
> > > the pauses occur I see kdeinit hit the top of the list and sometimes
> > > consuming 90% or more of a CPU and idle usage drops to 30-40%. I do
> > > see some latency traces (not really high ones) in the log that were
> > > generated by kdeinit but I think they were generated prior to when
> > > these pauses occurred, most likely when logging in.
> >
> > is this 90% or more CPU time system (kernel) overhead or userspace
> > overhead?

It appears that most of the time is being consumed in system (~50% vs. 
~6%).
> another thing to watch for is the context-switch rate in vmstat. -V0.3
> patches include a hack that include involuntary context-switches (mutex
> context switches) in the context-switch stat as well. (previously those
> were reported in a separate field which vmstat didnt pick up.)
>
> So if the context-switch rate shots up to above say 100K/sec that is a
> sure sign of some mutex badness. The livelock scenarios i solved in
> -V0.3 occasionally generated a more than 500K/sec context-switch rate on
> a 2GHz box. Having just a couple of thousand per sec isnt by itself a
> sign of anything unusual.

As for the context-switch, I do see this jump up to ~10-11K/sec from 
~4-5K/sec and I see this only when I trigger the pauses.

> 	Ingo

kr
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/