Ubuntu
linux package

Significant performance regression in 2.6.28-8.24 due to p4-clockmod

Bug #332017 reported by Max Bowsher
22
This bug affects 2 people
Affects Status Importance Assigned to Milestone
linux (Ubuntu)
Fix Released
High
Scott James Remnant (Canonical)

Bug Description

On upgrading to kernel 2.6.28-8.24, I noticed a very significant regression in the responsiveness of desktop applications: firefox became unreasonably slow to load pages, and screen redraws for things like switching tabs in xchat, gnome-terminal etc. went from being to instantaneous to taking a human-visible amount of time to process.

Adding the cpufreq panel applet revealed that my CPU was scaled back from 3GHz to 375MHz, and being excessively conservative about scaling up under load. Resetting to the performance governor cured the problem.

I suggest that the ondemand governor in some hardware may not be a viable default.

ProblemType: Bug
Architecture: amd64
DistroRelease: Ubuntu 9.04
MachineType: Supermicro P8SAA
NonfreeKernelModules: nvidia
Package: linux-image-2.6.28-8-generic 2.6.28-8.24
ProcCmdLine: root=LABEL=epsilon ro ht=on
ProcEnviron:
 LC_COLLATE=C
 PATH=(custom, user)
 LANG=en_GB.UTF-8
 SHELL=/bin/bash
ProcVersionSignature: Ubuntu 2.6.28-8.24-generic
SourcePackage: linux

Revision history for this message
Max Bowsher (maxb) wrote :
Revision history for this message
Max Bowsher (maxb) wrote :

The following may be in some way related, but not necessary complete duplicates:
bug 326149
bug 122993
bug 231856

Revision history for this message
Scott James Remnant (Canonical) (canonical-scott) wrote :

The ondemand governor has been the default for several releases now; the only thing that changed was the method we use to set it.

Changed in linux:
status: New → Invalid
Revision history for this message
Scott James Remnant (Canonical) (canonical-scott) wrote :

If you have a performance problem, please reopen the bug but change the description to not include words like "regression"

Revision history for this message
Scott James Remnant (Canonical) (canonical-scott) wrote :

Sorry, I was a little hasty with this one. Had far too many things on today.

As debugged on IRC, this regression is being caused by the use of the p4-clockmod scaling driver when previously you did not have one loaded.

This turns out to be a nasty, buggy, crufty driver that upstream kernel developers really don't recommend is used for frequency scaling.

I've comitted patches to not build in this driver, this may mean you're returned to having no frequency scaling.

Changed in linux:
assignee: nobody → scott
importance: Undecided → High
status: Invalid → Fix Committed
Revision history for this message
Max Bowsher (maxb) wrote :

This fixes the performance, and does indeed return to no frequency scaling.

Revision history for this message
tgpraveen (tgpraveen89) wrote :

i am on p4 2.4ghz processor with nvidia 6200 graphics card using propreitary drivers from nvidia.
today i upgraded from intrepid to jaunty and i too seem to be having the same problem i get the from the cpufreq panel applet that my cpu is set at 300 mhz and hence causes a lot of unresponsiveness.
so if i do

cpufreq-selector -g performance

then it solves the problem but cpu freq scaling is lost. pls solve this problem.

Revision history for this message
Tommaso R. Donnarumma (tawmas) wrote :

A new version of the kernel has been released, which includes Scott James Remnant's patch.

This fixes the issue for me, and others on the forums as well. I'm therefore closing this bug. Please, feel free to reopen it if you still have this problem.

Changed in linux:
status: Fix Committed → Fix Released
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.
Subscribing...

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.