lkml.org 
[lkml]   [2012]   [Jun]   [19]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRE: [PATCH] sched: Folding nohz load accounting more accurate
Date
> On 2012.06.18 09:04 -0700, Peter Zijlstra wrote:

[... lots deleted ...]

> Can someone please think through the below thing? its been compile
> tested only...

[... code patch deleted ...]

Your code patch makes sense to me, but I admit that I still have
difficulties to follow this area of code.

Note: I didn't have time yet to review or try Charles' solution.

I back edited this new solution into my working kernel and retested
the same two operating points as over the weekend. Summary:

2 Processes @ 90 hertz per process and 0.15 load per process,
or 0.30 total. Reported Load Average (long average):

Kernel 3.5 RC2: ~1.5
Kernel Peter 2012.06.15: ~1.8
Kernel Peter 2012.06.18: ~0.3 (0.28)

8 processes @ 150 hertz per process and 0.7925 load per process,
or 6.34 total. Reported Load Average (long average):

Kernel 3.5 RC2: ~3.9
Kernel Peter 2012.06.15: ~7.9
Kernel Peter 2012.06.18: ~6.3

I will start one of my longer term experiments tonight.
It will take many days to do all the tests.
If things change, the tests can be re-started.

Note: On my computers I have no way to test the catch-
up code path, as my computers never take that path.

A note on the test code for loading (from other branch
of this thread): Peter, I'll try your code sometime. It was
on purpose that I made mine a mindless code loop, without any
system calls to keep time. But yes, mine is proving a little
annoying to use.





\
 
 \ /
  Last update: 2012-06-19 09:01    [W:0.191 / U:0.396 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site