Re: [git pull] scheduler changes for v2.6.26

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Tue, 2008-04-22 at 10:59 +0200, Ingo Molnar wrote:
> * Mike Galbraith <efault@gmx.de> wrote:
> 
> > > the 700-800 msecs of delays you see are very "brutal" so there must 
> > > be something fundamentally wrong going on here.
> > 
> > I'm seeing latency hits with 26.git, whereas 25 is hit free.
> 
> ok. I think what happens is that your broken sched-clock hid the real 
> breakage. Lets try fix the real breakage now.
> 
> I've uploaded a new sched-devel.git that is against very latest -git, 
> could you try ftrace (with a sufficiently large 
> /debug/tracing/tracing_max_entries value) - perhaps the 
> worst-case-wakeup-latency tracer shows large latencies? If not, then 
> maybe the sched_switch tracer gives a better insight into what's 
> happening?
> 
> > Erm, should my Q6600 emit such?
> 
> on nohz it could happen - and fixed in -git. Patch looked too dangerous 
> for late-2.6.25 to merge.

Also, it only happens when the cpu has idle time; and that typically
happens when its well,. idle - so not much to schedule wrong.

That said; yes there are boundary effect that could make it show up.

> > On 26.git, I get numbers like yours, but with occasional dips down to 
> > ~700, though the latency hits don't _seem_ to be synchronous with 
> > watch-rq-clock.sh glitchies.
> 
> hm, the dips shouldnt be happening normally.

Agreed, those ought to be gone.. Guillaume do you see any holes in the
current rq clock code?