lkml.org 
[lkml]   [2010]   [Sep]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [RFC patch 1/2] sched: dynamically adapt granularity with nr_running
From
Date
On Sat, 2010-09-11 at 13:48 -0700, Linus Torvalds wrote:
> On Sat, Sep 11, 2010 at 1:36 PM, Peter Zijlstra <peterz@infradead.org> wrote:
> >
> >From what I can make up:
> >
> > LAT=`cat /proc/sys/kernel/sched_latency_ns`;
> > echo $((LAT/8)) > /proc/sys/kernel/sched_min_granularity_ns
> >
> > will give you pretty much the same result as Mathieu's patch.
>
> Or perhaps not. The point being that Mathieu's patch seems to do this
> dynamically based on number of runnable threads per cpu. Which seems
> to be a good idea.
>
> IOW, this part:
>
> - if (delta_exec < sysctl_sched_min_granularity)
> + if (delta_exec < __sched_gran(cfs_rq->nr_running))
>
> seems to be a rather fundamental change, and looks at least
> potentially interesting. It seems to make conceptual sense to take the
> number of running tasks into account at that point, no?

We used to have something like that a long while back, we nixed it
because of the division and replaced it with floor(__sched_gran) (ie.
the smallest value it would ever give).

Smaller values are better for latency, larger values are better for
throughput. So introducing __sched_gran() in order to provide larger
values doesn't make sense to me.

> And I don't like how you dismissed the measured latency improvement.
> And yes, I do think latency matters. A _lot_.

OK, we'll make it better and sacrifice some throughput, can do, no
problem.

> And no, I'm not saying that Mathieu's patch is necessarily good. I
> haven't tried it myself. I don't have _that_ kind of opinion. The
> opinion I do have is that I think it's sad how you dismissed things
> out of hand - and seem to _continue_ to dismiss them without
> apparently actually having looked at the patch at all.

Let me draw you a picture of what this patch looks like to me:

* is slice length, + is period length

Patch (sched_latency = 10, sched_min_gran = 10/3)


30 | +
|
|
| +
|
|
|
|
|
|
20 |
|
|
|
|
|
|
|
|
|
10 | * + + + + + + +
|
|
|
|
| *
|
| * * * * * * * *
| * *
| * *
0 +---------------------------------------------------------
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16


Normal (sched_latency = 10, sched_min_gran = 10/3)


30 | +
|
|
| +
|
|
|
| +
|
|
20 | +
|
|
| +
|
|
|
| +
|
|
10 | * + +
|
|
|
|
| *
|
| * * * * * * * * * * * * * *
|
|
0 +---------------------------------------------------------
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16



Normal (sched_latency = 10, sched_min_gran = 10/8)

30 |
|
|
|
|
|
|
|
|
|
20 |
|
|
|
|
| +
| +
| +
|
| +
10 | * + + + + + + +
|
|
|
|
| *
|
| * *
| * *
| * * * * * * * *
0 +---------------------------------------------------------
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16













\
 
 \ /
  Last update: 2010-09-12 11:09    [W:0.159 / U:0.236 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site