Re: [v7 0/8] Reduce cross CPU IPI interference

From: Gilad Ben-Yossef
Date: Thu Feb 02 2012 - 03:46:30 EST

On Wed, Feb 1, 2012 at 7:04 PM, Frederic Weisbecker <fweisbec@xxxxxxxxx> wrote:
> On Sun, Jan 29, 2012 at 10:25:46AM +0200, Gilad Ben-Yossef wrote:
> > If this is of interest, I keep a list tracking global IPI and global
> > task schedulers sources in the core kernel here:
> >
> >
> > I plan to visit all these potential interference source to see if
> > something can be done to lower their effect on
> > isolated CPUs over time.
> Very nice especially as many people seem to be interested in
> CPU isolation.

Yes, that is what drives me as well. I have a bare metal program
I'm trying to kill here, I researched CPU isolation and ran into your
nohz patch set and asked myself: "OK, if we disable the tick what else
is on the way?"

> When we get the adaptive tickless feature in place, perhaps we'll
> also need to think about some way to have more control on the
> CPU affinity of some non pinned timers to avoid disturbing
> adaptive tickless CPUs. We still need to consider their cache affinity
> though.

Right. I'm thinking we can treat a CPU going in adaptive tick mode in a similar
fashion to a CPU going offline for the purpose of timer migration.

Some pinned timers might be able to get special treatment as well - take for
example the vmstat work being schedule every second, what should we do with
it for CPU isolation?

It makes sense to me to have that stop scheduling itself when we have the tick
disabled for both idle and a nohz task.

A similar thing can be said for the clocksource watchdog for example - we might
consider having it not trigger stuff on idle or nohz task CPUs

Maybe we can have some notification mechanism when a task goes into nohz
mode and back to let stuff disable itself and back if it makes sense.
It seems more
sensible then having all these individual pieces check for whether
this CPU or other is
in idle or nohz task mode.

The question for nohz task then is when does the notification needs to go out?
only when a task managed to go into nohz mode or when we add a cpu to an
adaptive tick cpuset? because for stuff like vmstat, the very existence of the
runnable workqueue thread can keep a task from going into nohz mode. bah.
maybe we need two notifications...

Gilad Ben-Yossef
Chief Coffee Drinker
Israel Cell: +972-52-8260388
US Cell: +1-973-8260388

"If you take a class in large-scale robotics, can you end up in a
situation where the homework eats your dog?"
 -- Jean-Baptiste Queru
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at