lkml.org 
[lkml]   [2014]   [Jan]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC PATCH] sched: find the latest idle cpu
On 01/15/2014 01:33 PM, Michael wang wrote:
> On 01/15/2014 12:07 PM, Alex Shi wrote:
>> > Currently we just try to find least load cpu. If some cpus idled,
>> > we just pick the first cpu in cpu mask.
>> >
>> > In fact we can get the interrupted idle cpu or the latest idled cpu,
>> > then we may get the benefit from both latency and power.
>> > The selected cpu maybe not the best, since other cpu may be interrupted
>> > during our selecting. But be captious costs too much.
> So the idea here is we want to choose the latest idle cpu if we have
> multiple idle cpu for choosing, correct?

yes.
>
> And I guess that was in order to avoid choosing tickless cpu while there
> are un-tickless idle one, is that right?

no, current logical choice least load cpu no matter if it is idle.
>
> What confused me is, what about those cpu who just going to recover from
> tickless as you mentioned, which means latest idle doesn't mean the best
> choice, or even could be the worst (if just two choice, and the longer
> tickless one is just going to recover while the latest is going to
> tickless).

yes, to save your scenario, we need to know the next timer for idle cpu,
but that is not enough, interrupt is totally unpredictable. So, I'd
rather bear the coarse method now.
>
> So what about just check 'ts->tick_stopped' and record one ticking idle
> cpu? the cost could be lower than time comparison, we could reduce the
> risk may be...(well, not so risky since the logical only works when
> system is relaxing with several cpu idle)

first, nohz full also stop tick. second, tick_stopped can not reflect
the interrupt. when the idle cpu was interrupted, it's waken, then be a
good candidate for task running.

--
Thanks
Alex


\
 
 \ /
  Last update: 2014-01-15 08:21    [W:0.107 / U:2.120 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site