Arch Linux

Please read this before reporting a bug:
https://wiki.archlinux.org/title/Bug_reporting_guidelines

Do NOT report bugs when a package is just outdated, or it is in the AUR. Use the 'flag out of date' link on the package page, or the Mailing List.

REPEAT: Do NOT report bugs for outdated packages!
Tasklist

FS#16152 - [kernel26] improved scheduler behavior for desktops

Attached to Project: Arch Linux
Opened by Pawel (kraftman) - Saturday, 12 September 2009, 09:04 GMT
Last edited by Roman Kyrylych (Romashka) - Saturday, 03 October 2009, 19:32 GMT
Task Type Bug Report
Category Kernel
Status Closed
Assigned To Tobias Powalowski (tpowa)
Aaron Griffin (phrakture)
Architecture All
Severity Low
Priority Normal
Reported Version
Due in Version Undecided
Due Date Undecided
Percent Complete 100%
Votes 1
Private No

Details

Hi,

There was quite interesting discussion going on lkml about scheduler latency and performance. Ingo discovered (with some people help) there's a bug in NEW_FAIR_SLEEPERS scheduler feature which can cause visible latency for some users.

It's described here:

before change:

http://lkml.org/lkml/2009/9/10/53


8.529265 task-clock-msecs # 0.001 CPUs
23 context-switches # 0.003 M/sec
1 CPU-migrations # 0.000 M/sec
315 page-faults # 0.037 M/sec

< 11.804293482 seconds time elapsed >

and after disabling NEW_FAIR_SLEEPERS:

http://lkml.org/lkml/2009/9/10/229


9.009137 task-clock-msecs # 0.447 CPUs
18 context-switches # 0.002 M/sec
1 CPU-migrations # 0.000 M/sec
315 page-faults # 0.035 M/sec

< 0.020167093 seconds time elapsed >

AFAIK NEW_FAIR_SLEEPERS are disabled for 2.6.32, but maybe it's a good idea to have them disabled now? Rather shouldn't introduce problems. Easy way to turn them of:

http://www.phoronix.com/forums/showpost.php?p=91609&postcount=40

P.S. Kudos to Nikos who reported this :>
This task depends upon

Closed by  Roman Kyrylych (Romashka)
Saturday, 03 October 2009, 19:32 GMT
Reason for closing:  Deferred
Additional comments about closing:  we will wait for upstream change
Comment by Pawel (kraftman) - Saturday, 12 September 2009, 20:35 GMT
It also resolves some more problems which are described in quite long thread at lkml ;)
Comment by Tobias Powalowski (tpowa) - Sunday, 13 September 2009, 15:13 GMT
At the moment no code has been taken to .32 series.
As the statement says, disable it for now until the real bug is found.
Well, until it is not merged upstream i don't see a reason to add this to our kernel.
Comment by Pawel (kraftman) - Sunday, 13 September 2009, 16:36 GMT
Yes, there's only request to pull this and it's not in the mainline yet.
Comment by Pawel (kraftman) - Tuesday, 22 September 2009, 10:01 GMT
I think it is good idea to close this report and just stick to the defaults. Thanks.

Loading...