Re: [PATCH v3 2/5] sched/deadline: Fix reclaim inaccuracy with SMP

[luca abeni]

On Mon, 15 May 2023 21:47:03 -0400
Vineeth Remanan Pillai <vineeth@xxxxxxxxxxxxxxx> wrote:

> Hi Luca,
> 
> On Mon, May 15, 2023 at 4:06 AM luca abeni
> <luca.abeni@xxxxxxxxxxxxxxx> wrote:
> 
> >
> > this patch is giving me some headaches:
> >  
> Sorry about that.. I was also stressing out on how to get the
> reclaiming done right for the past couple of days ;-)

Well, this math is hard... :)

> > Vineeth Pillai <vineeth@xxxxxxxxxxxxxxx> wrote:
> > [...]  
> > >   *   Uextra:         Extra bandwidth not reserved:
> > > - *                   = Umax - \Sum(u_i / #cpus in the root
> > > domain)
> > > + *                   = Umax - this_bw  
> >
> > While I agree that this setting should be OK, it ends up with
> >         dq = -Uact / Umax * dt
> > which I remember I originally tried, and gave some issues
> > (I do not remember the details, but I think if you try N
> > identical reclaiming tasks, with N > M, the reclaimed time
> > is not distributed equally among them?)
> >  
> I have noticed this behaviour where the reclaimed time is not equally
> distributed when we have more tasks than available processors. But it
> depended on where the task was scheduled. Within the same cpu, the
> distribution seemed to be proportional.

Yes, as far as I remember it is due to migrations. IIRC, the problem is
related to the fact that using "dq = -Uact / Umax * dt" a task running
on a core might end up trying to reclaim some idle time from other
cores (which is obviously not possible).
This is why m-GRUB used "1 - Uinact" instead of "Uact"

[...]
> > I need to think a little bit more about this...
> >  
> Thanks for looking into this.. I have a basic idea why tasks with less
> bandwidth reclaim less in SMP when number of tasks is less than number
> of cpus, but do not yet have a verifiable fix for it.

I think I can now understand at least part of the problem. In my
understanding, the problem is due to using
	dq = -(max{u_i, (Umax - Uinact - Uextra)} / Umax) * dt

It should really be
	dq = -(max{u_i, (1 - Uinact - Uextra)} / Umax) * dt

(since we divide by Umax, using "Umax - ..." will lead to reclaiming up
to "Umax / Umax" = 1)

Did you try this equation?

I'll write more about this later... And thanks for coping with all my
comments!


				Luca
> 
> If patches 1 and 4 looks good to you, we shall drop 2 and 3 and fix
> the SMP issue with varying bandwidth separately.. Patch 4 would
> differ a bit when I remove 2 and 3 so as to use the formula:
>  "dq = -(max{u, (Umax_reclaim - Uinact - Uextra)} / Umax_reclaim) dt"
> 
> Thanks for your patience with all these brainstorming:-)
> 
> Thanks,
> Vineeth