[lkml]   [2011]   [Jan]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: 'autogroup' sched code KILLING responsiveness
> Christian, Michael, can you confirm that this and the second patch fixes the 
> interactivity bug for you? If yes then i'd like to apply and send this fix to Linus
> Thanks,
> Ingo

I'm currently running with autogroup enabled for the first time,
on Athlon XP -rc2 with _first patch only_ applied (saw second part too late),
and it appears to be working fine with some stress tests added
(multi-app benchmark, while... true, yes etc.).
Interestingly I had chosen to DISABLE the cgroup part of the scheduler
several months ago since it was slower than non-cgroup by quite some margin.

One should perhaps consider this issue to be evidence of a notable lack
of (possibly automated) test/usage coverage of UP configs
(together with the recent build error on UP that was discovered
once hitting -rc1 only, and not earlier!).

For this particular fix, perhaps the 0 vs. tg->shares change could be
followed up by a WARN_ON_ONCE() in some more internal non-hotpath function
if such a check is feasible in the current case.


Andreas Mohr

 \ /
  Last update: 2011-01-23 19:55    [W:0.068 / U:0.104 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site