Re: [RFC v3 1/6] Track the active utilisation

[luca abeni]

Hi Juri,

On Tue, 1 Nov 2016 16:45:43 +0000
Juri Lelli <juri.lelli@xxxxxxx> wrote:

> Hi,
> 
> a few nitpicks on subject and changelog and a couple of questions below.
> 
> Subject should be changed to something like
> 
>  sched/deadline: track the active utilisation
Ok; that's easy :)
I guess a similar change should be applied to the subjects of all the
other patches, right?


> 
> On 24/10/16 16:06, Luca Abeni wrote:
> > The active utilisation here is defined as the total utilisation of the  
> 
> s/The active/Active/
> s/here//
> s/of the active/of active/
Ok; I'll do this in the next revision of the patchset.


> > active (TASK_RUNNING) tasks queued on a runqueue. Hence, it is increased
> > when a task wakes up and is decreased when a task blocks.
> > 
> > When a task is migrated from CPUi to CPUj, immediately subtract the task's
> > utilisation from CPUi and add it to CPUj. This mechanism is implemented by
> > modifying the pull and push functions.
> > Note: this is not fully correct from the theoretical point of view
> > (the utilisation should be removed from CPUi only at the 0 lag time),  
> 
> a more theoretically sound solution will follow.
Notice that even the next patch (introducing the "inactive timer") ends up
migrating the utilisation immediately (on tasks' migration), without waiting
for the 0-lag time.
This is because of the reason explained in the following paragraph:

> > but doing the right thing would be _MUCH_ more complex (leaving the
> > timer armed when the task is on a different CPU... Inactive timers should
> > be moved from per-task timers to per-runqueue lists of timers! Bah...)  
> 
> I'd remove this paragraph above.
Ok. Re-reading the changelog, I suspect this is not the correct place for this
comment.


> > The utilisation tracking mechanism implemented in this commit can be
> > fixed / improved by decreasing the active utilisation at the so-called
> > "0-lag time" instead of when the task blocks.  
> 
> And maybe this as well, or put it as more information about the "more
> theoretically sound" solution?
Ok... I can remove the paragraph, or point to the next commit (which
implements the more theoretically sound solution). Is such a "forward
reference" in changelogs ok?

[...]
> > @@ -947,14 +965,19 @@ static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags)
> >  		return;
> >  	}
> >  
> > +	if (p->on_rq == TASK_ON_RQ_MIGRATING)
> > +		add_running_bw(&p->dl, &rq->dl);
> > +
> >  	/*
> >  	 * If p is throttled, we do nothing. In fact, if it exhausted
> >  	 * its budget it needs a replenishment and, since it now is on
> >  	 * its rq, the bandwidth timer callback (which clearly has not
> >  	 * run yet) will take care of this.
> >  	 */
> > -	if (p->dl.dl_throttled && !(flags & ENQUEUE_REPLENISH))
> > +	if (p->dl.dl_throttled && !(flags & ENQUEUE_REPLENISH)) {
> > +		add_running_bw(&p->dl, &rq->dl);  
> 
> Don't rememeber if we discussed this already, but do we need to add the bw here
> even if the task is not actually enqueued until after the replenishment timer
> fires?
I think yes... The active utilization does not depend on the fact that the task
is on the runqueue or not, but depends on the task's state (in GRUB parlance,
"inactive" vs "active contending"). In other words, even when a task is throttled
its utilization must be counted in the active utilization.


[...]
> >  	/*
> >  	 * Since this might be the only -deadline task on the rq,
> >  	 * this is the right place to try to pull some other one
> > @@ -1712,6 +1748,7 @@ static void switched_from_dl(struct rq *rq, struct task_struct *p)
> >   */
> >  static void switched_to_dl(struct rq *rq, struct task_struct *p)
> >  {
> > +	add_running_bw(&p->dl, &rq->dl);
> >  
> >  	/* If p is not queued we will update its parameters at next wakeup. */
> >  	if (!task_on_rq_queued(p))  
> 
> Don't we also need to remove bw in task_dead_dl()?
I think task_dead_dl() is invoked after invoking dequeue_task_dl(), which takes care
of this... Or am I wrong? (I think I explicitly tested this, and modifications to
task_dead_dl() turned out to be unneeded)



			Thanks,
				Luca