On Tue, 2012-09-11 at 11:55 -0700, Sven-Thorsten Dietrich wrote: 

On Sep 11, 2012, at 11:10 AM, Mike Galbraith [off-list ref] wrote:

quoted

On Tue, 2012-09-11 at 17:34 +0200, Wolfgang Wallner wrote:

quoted

quoted

quoted

My questions are now:

* What does this logging entry mean?
Could you please point me to some information about RT throttling
so that I can understand what's it about?

quoted

With stock settings, it means realtime task[s] consumed > 95% of the
throttle interval (1s), so the throttle activated, allowing

SCHED_NORMAL

quoted

tasks to have a sip of CPU, to let you try to save the box from

nutty RT

quoted

CPU hogs. See kernel/sched_rt.c.

I think if the application turns into a cpu hog...

Where from comes 'if'?  You presented evidence, so methinks there's not
a _lot_ of room for an 'if', there's just a missing 'why'.

Well, as you know a lot of folks do polling, but its important to make
sure not ALL cores are doing that...

[60382.945209] EplTimerHighResk: Continuous timer (handle 0x10000001)
had to skip 836 interval(s)!

That and 'stops working' (as in forever) made me suspect spinner rather
than transient cpu over-commit.

Need to look at the code, maybe we could add a mask to restrict
throttling per cpu (e.g. to 0) ?

IMHO it's a debugging tool you turn off for normal operation.  Biggest
problem I've ever had with the thing is it allowing hogs to run over to
the neighbors to borrow a cup of CPU.  That's fixed.

-Mike