Re: [RFC PATCH v2 6/6] sched/fair: Implement starvation monitor

From: luca abeni
Date: Fri Aug 07 2020 - 09:50:38 EST

Hi Peter,

On Fri, 7 Aug 2020 12:46:18 +0200
peterz@xxxxxxxxxxxxx wrote:

> On Fri, Aug 07, 2020 at 11:56:04AM +0200, Juri Lelli wrote:
> > Starting deadline server for lower priority classes right away when
> > first task is enqueued might break guarantees, as tasks belonging to
> > intermediate priority classes could be uselessly preempted. E.g., a
> > well behaving (non hog) FIFO task can be preempted by NORMAL tasks
> > even if there are still CPU cycles available for NORMAL tasks to
> > run, as they'll be running inside the fair deadline server for some
> > period of time.
> >
> > To prevent this issue, implement a starvation monitor mechanism that
> > starts the deadline server only if a (fair in this case) task hasn't
> > been scheduled for some interval of time after it has been enqueued.
> > Use pick/put functions to manage starvation monitor status.
> One thing I considerd was scheduling this as a least-laxity entity --
> such that it runs late, not early

Are you thinking about scheduling both RT and non-RT tasks through
deadline servers? If yes, then I think that using something like
laxity-based scheduling for the SCHED_OTHER server can be a good idea
(but then we need to understand how to combine deadline-based
scheduling with laxity-based scheduling, etc...)

Or are you thinking about keeping the SCHED_OTHER server throttled
until its laxity is 0 (or until its laxity is lower than some small
value)? In this second case, the approach would work even if RT tasks
are not scheduled through a server (but I do not know which kind of
performance guarantee we could provide).

> -- and start the server when
> rq->nr_running != rq->cfs.h_nr_running, IOW when there's !fair tasks
> around.

Yes, this could be a good optimization.

> Not saying we should do it like that, but that's perhaps more
> deterministic than this.