lkml.org 
[lkml]   [2014]   [Apr]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 2/2] nohz: use delayed iowait accounting to avoid race on idle time stats
On Fri, Apr 18, 2014 at 02:52:05PM +0900, Hidetoshi Seto wrote:
> I think it also works... but I have some concerns here:
>
> - it changes golden path in scheduler core.
> impact for performance is questionable.

We should be able to measure that, but like said, it shouldn't change
the number of atomic ops and makes sure its all in the same cacheline,
so the additional code is all cheap.

> - it forces managing last_iowait even if system is in busy
> I guess it will drop max performance of the system
> while my proposed fix only touches procedure for idle
> with nohz.

That could be fixed by looking to see if the remote rq is idle.

> By the way, I have posted my v4 patch set:
> https://lkml.org/lkml/2014/4/17/120
>
> I'll happy if you could give your comments on it too!

Yeah, I saw that, didn't have time yet, will hopefully get to it soon
:-)


\
 
 \ /
  Last update: 2014-04-18 11:21    [W:0.069 / U:0.564 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site