← Back to team overview

kernel-packages team mailing list archive

[Bug 1219548] [NEW] sched_autogroup_enabled breaks niceness

 

Public bug reported:

With the default Ubuntu 13.04 sysctl settings and boot parameters,
positively niced tasks are very clearly given a larger share of CPU time
than they ought to be. The update-apt-xapian-index process, for
instance, runs at niceness +19; but it still slows my Celeron M520
laptop to a crawl. Likewise for other positively niced tasks.

If, on the other hand, I boot with the 'noautogroup' kernel parameter
(which sets kernel.sched_autogroup_enabled to 0), responsiveness is much
better when running CPU heavy tasks at positive niceness.

In both cases, processes are reported by top and ps as running at the
correct niceness; it appears that the nice level is ignored silently.

Autogrouping is supposed to provide better responsiveness under some
loads, but I believe that breaking nice settings is too high a price. I
think it would be a good idea to turn off
kernel.sched_autogroup_enabled, until this behavior is fixed upstream.

By the way, an important note: modifying the sched_autogroup_enabled
sysctl on a running system will usually cause a kernel panic.
Autogrouping needs to be turned off using 'noautogroup' as a boot
parameter; if you try it using sysctl.conf you will get a nice kernel
panic on boot.

** Affects: linux (Ubuntu)
     Importance: Undecided
         Status: New

-- 
You received this bug notification because you are a member of Kernel
Packages, which is subscribed to linux in Ubuntu.
https://bugs.launchpad.net/bugs/1219548

Title:
  sched_autogroup_enabled breaks niceness

Status in “linux” package in Ubuntu:
  New

Bug description:
  With the default Ubuntu 13.04 sysctl settings and boot parameters,
  positively niced tasks are very clearly given a larger share of CPU
  time than they ought to be. The update-apt-xapian-index process, for
  instance, runs at niceness +19; but it still slows my Celeron M520
  laptop to a crawl. Likewise for other positively niced tasks.

  If, on the other hand, I boot with the 'noautogroup' kernel parameter
  (which sets kernel.sched_autogroup_enabled to 0), responsiveness is
  much better when running CPU heavy tasks at positive niceness.

  In both cases, processes are reported by top and ps as running at the
  correct niceness; it appears that the nice level is ignored silently.

  Autogrouping is supposed to provide better responsiveness under some
  loads, but I believe that breaking nice settings is too high a price.
  I think it would be a good idea to turn off
  kernel.sched_autogroup_enabled, until this behavior is fixed upstream.

  By the way, an important note: modifying the sched_autogroup_enabled
  sysctl on a running system will usually cause a kernel panic.
  Autogrouping needs to be turned off using 'noautogroup' as a boot
  parameter; if you try it using sysctl.conf you will get a nice kernel
  panic on boot.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1219548/+subscriptions


Follow ups

References