Re: [PATCH 2/2] sched: adjust SCHED_IDLE interactions

[Vincent Guittot]

On Thu, 12 Aug 2021 at 23:09, Josh Don <joshdon@xxxxxxxxxx> wrote:
>
> > > @@ -697,8 +699,18 @@ static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
> > >                 slice = __calc_delta(slice, se->load.weight, load);
> > >         }
> > >
> > > -       if (sched_feat(BASE_SLICE))
> > > -               slice = max(slice, (u64)w);
> > > +       if (sched_feat(BASE_SLICE)) {
> > > +               /*
> > > +                * SCHED_IDLE entities are not subject to min_granularity if
> > > +                * they are competing with non SCHED_IDLE entities. As a result,
> > > +                * non SCHED_IDLE entities will have reduced latency to get back
> > > +                * on cpu, at the cost of increased context switch frequency of
> > > +                * SCHED_IDLE entities.
> > > +                */
> >
> > Ensuring that the entity will have a minimum runtime has been added to
> > ensure that we let enough time to move forward.
> > If you exclude sched_idle entities from this min runtime, the
> > sched_slice of an idle_entity will be really small.
> > I don't have details of your example above but I can imagine that it's
> > a 16 cpus system which means a sysctl_sched_min_granularity=3.75ms
> > which explains the 4ms running time of an idle entity
> > For a 16 cpus system, the sched_slice of an idle_entity in your
> > example in the cover letter is: 6*(1+log2(16))*3/1027=87us. Of course
> > this become even worse with more threads and cgroups or thread with
> > ncie prio -19
> >
> > This value is then used to set the next hrtimer event in SCHED_HRTICK
> > and 87us is too small to make any progress
> >
> > The 1ms of your test comes from the tick which could be a good
> > candidate for a min value or the
> > normalized_sysctl_sched_min_granularity which has the advantage of not
> > increasing with number of CPU
>
> Fair point, this shouldn't completely ignore min granularity. Something like
>
> unsigned int sysctl_sched_idle_min_granularity = NSEC_PER_MSEC;
>
> (and still only using this value instead of the default
> min_granularity when the SCHED_IDLE entity is competing with normal
> entities)

Yes that looks like a good option

Also note that with a NSEC_PER_MSEC default value, the sched_idle
entity will most probably run 2 ticks instead of the 1 tick (HZ=1000)
that you have with your proposal because a bit less than a full tick
is accounted to the running thread (the time spent in interrupt is not
accounted as an example) so sysctl_sched_idle_min_granularity of 1ms
with HZ=1000 will most propably run 2 ticks. Instead you could reuse
the default 750000ULL value of sched_idle_min_granularity

That being said sysctl_sched_idle_min_granularity =
normalized_sysctl_sched_min_granularity * scale_factor which means
that normalized_sysctl_sched_min_granularity stays the same
(750000ULL) whatever the number of cpus

>
> > > @@ -4216,7 +4228,15 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
> > >                 if (sched_feat(GENTLE_FAIR_SLEEPERS))
> > >                         thresh >>= 1;
> > >
> > > -               vruntime -= thresh;
> > > +               /*
> > > +                * Don't give sleep credit to a SCHED_IDLE entity if we're
> > > +                * placing it onto a cfs_rq with non SCHED_IDLE entities.
> > > +                */
> > > +               if (!se_is_idle(se) ||
> > > +                   cfs_rq->h_nr_running == cfs_rq->idle_h_nr_running)
> >
> > Can't this condition above create unfairness between idle entities ?
> > idle thread 1 wake up while normal thread is running
> > normal thread thread sleeps immediately after
> > idle thread 2 wakes up just after and gets some credits compared to the 1st one.
>
> Yes, this sacrifices some idle<->idle fairness when there is a normal
> thread that comes and goes. One alternative is to simply further
> reduce thresh for idle entities. That will interfere with idle<->idle
> fairness when there are no normal threads, which is why I opted for
> the former. On second thought though, the former fairness issue seems
> more problematic. Thoughts on applying a smaller sleep credit
> threshold universally to idle entities?

This one is a bit more complex to set.
With adding 1, you favor the already runnable tasks by ensuring that
they have or will run a slice during this period before sched_idle
task
But as soon as you subtract something to min_vruntime, the task will
most probably be scheduled at the next tick if other tasks already run
for a while (at least a sched period). If we use
sysctl_sched_min_granularity for sched_idle tasks that wake up instead
of sysctl_sched_latency, we will ensure that a sched_idle task will
not preempt a normal task, which woke up few ms before, and we will
keep some fairness for sched_idle task that sleeps compare to other.

so a thresh of sysctl_sched_min_granularity (3.75ms with 16 cpus )
should not disturb your UC and keep some benefit for newly wake up
sched_ide task