PDA

View Full Version : measuring desktop responsiveness?

KhaaL
04-20-2009, 03:39 PM
I'm one of many who are affected by kernel bug 12309 ( http://bugzilla.kernel.org/show_bug.cgi?id=12309 ) and as you can see, the major symptom is that desktop responsivness taking a hit during heavy I/O operations. Thing is, there is no way to measure desktop responsiveness objectively that I know of. Could it be possible to see PTS to include such a test in the future? it would be quite helpful and useful for future usage i'm sure :)
Michael
04-22-2009, 02:25 PM
I have an idea for possibly making an automated test of this... Will check when I am back in the office.

Michael
kraftman
04-22-2009, 04:33 PM
I'm one of many who are affected by kernel bug 12309 ( http://bugzilla.kernel.org/show_bug.cgi?id=12309 ) and as you can see, the major symptom is that desktop responsivness taking a hit during heavy I/O operations. Thing is, there is no way to measure desktop responsiveness objectively that I know of. Could it be possible to see PTS to include such a test in the future? it would be quite helpful and useful for future usage i'm sure :)

Maybe you should try 2.6.30-rc3, because there are "some block IO scheduling fixes". Btw. is this bug confirmed by kernel devs? As I remember there was something about this or similar issue at lkml, but it wasn't really bug. What fs are you using? You suffer from this bug doing what operations? Are you using FUSE?
KhaaL
04-23-2009, 05:14 AM
I have an idea for possibly making an automated test of this... Will check when I am back in the office.

Michael

Great, I'm looking forward to try this out!

Maybe you should try 2.6.30-rc3, because there are "some block IO scheduling fixes". Btw. is this bug confirmed by kernel devs? As I remember there was something about this or similar issue at lkml, but it wasn't really bug. What fs are you using? You suffer from this bug doing what operations? Are you using FUSE?

Thanks for the suggestions. This bug is hard to confirm by the devs - or rather its hard to confirm the cause of the bug. so far I've experienced it on ext3, ext4 and reiserFS without any use of FUSE. I'll try .30-rc3 as soon nvidia release a driver compatible with it :-)
Kano
04-23-2009, 05:34 AM
Nvidia 180.53 works with 2.6.30.
ebird
04-23-2009, 11:35 AM
There is a test case which makes it possible to measure the desktop responsiveness latencies.

See comment:
http://bugzilla.kernel.org/show_bug.cgi?id=12309#c311

Test case:
http://bugzilla.kernel.org/attachment.cgi?id=21054

The problem is, that this bug is not deterministic. You can create heavy io without notice the bug, but mostly there is the poor desktop responsiveness.

See the three test results of 2.6.29 or 2.6.30-rc2-smp in the result log.
http://bugzilla.kernel.org/attachment.cgi?id=21055
grigi
04-23-2009, 12:28 PM
I have the same issue with a box using XFS, I found it kind-of unexplainable, but going though that bug it makes sense now :)
kraftman
04-23-2009, 01:31 PM
I have the same issue with a box using XFS, I found it kind-of unexplainable, but going though that bug it makes sense now :)

Maybe only CFQ is affected? Does someone tried with another scheduler? And maybe it's already fixed in 2.6.30-rc3:

http://bugzilla.kernel.org/show_bug.cgi?id=12309#c313
ebird
04-23-2009, 02:34 PM
This bug exists since 2.6.18 or even earlier. The real cause is unknowns. There were many tries to fix this bug in ext3 implementation, block layer, io scheduler or event cpu scheduler, without real improvements.

bug in bugzilla (http://bugzilla.kernel.org/show_bug.cgi?id=12309)

bug in ubuntu's launchpad (https://bugs.launchpad.net/ubuntu/+source/linux/+bug/131094)