Audio skipping with 2.6.24.4 and 2.6.25-rc9
I'm experiencing a lot of problems with audio skipping on my system. I am using Rhythmbox as an audio player and have Compiz active. Whenever I maximise or restore a maximised real transparent gnome-terminal, audio skips. Compared to other normal applications, restoring the transparent console is also slow (it takes a few seconds before the window appears).
I ran cfs-debug-info-clear.sh, and the reproduced the problem 3 times by restoring and minimising a gnome-terminal window. After that I ran cfs- debug-info.sh, and I uploaded the information here:
http://artipc10.vub.ac.be/cfs-debug-info-2008.04.12-18.00.20.gz
This problem happens with both 2.6.24.4 as with 2.6.25-rc9 (the debugging information is from the latter version).
Signature · 1 line
Frederik Himpe
Quoted 9 lines · Frederik Himpe
I'm experiencing a lot of problems with audio skipping on my system. I am using Rhythmbox as an audio player and have Compiz active. Whenever I maximise or restore a maximised real transparent gnome-terminal, audio skips. Compared to other normal applications, restoring the transparent console is also slow (it takes a few seconds before the window appears).
...
This problem happens with both 2.6.24.4 as with 2.6.25-rc9 (the debugging information is from the latter version).
Try to disable the group scheduler in your kernel configuration. It solved the audio skipping problem on my machine...
Signature · 1 line
Markus
Quoted 11 lines · Markus Trippelsdorf
Quoted 10 lines · Frederik Himpe
I'm experiencing a lot of problems with audio skipping on my system. I am using Rhythmbox as an audio player and have Compiz active. Whenever I maximise or restore a maximised real transparent gnome-terminal, audio skips. Compared to other normal applications, restoring the transparent console is also slow (it takes a few seconds before the window appears). ...
This problem happens with both 2.6.24.4 as with 2.6.25-rc9 (the debugging information is from the latter version).
Quoted 2 lines
Try to disable the group scheduler in your kernel configuration. It solved the audio skipping problem on my machine...
I think I have it disabled already?
$ zcat /proc/config.gz | grep GROUP # CONFIG_CGROUPS is not set # CONFIG_GROUP_SCHED is not set
Or am I missing something?
Signature · 1 line
Frederik Himpe
Quoted 16 lines · Frederik Himpe
Quoted 11 lines · Markus Trippelsdorf
Quoted 10 lines · Frederik Himpe
I'm experiencing a lot of problems with audio skipping on my system. I am using Rhythmbox as an audio player and have Compiz active. Whenever I maximise or restore a maximised real transparent gnome-terminal, audio skips. Compared to other normal applications, restoring the transparent console is also slow (it takes a few seconds before the window appears). ...
This problem happens with both 2.6.24.4 as with 2.6.25-rc9 (the debugging information is from the latter version).Quoted 2 lines
Try to disable the group scheduler in your kernel configuration. It solved the audio skipping problem on my machine...
Quoted 7 lines
I think I have it disabled already?
$ zcat /proc/config.gz | grep GROUP # CONFIG_CGROUPS is not set # CONFIG_GROUP_SCHED is not setOr am I missing something?
Yeah, I figured that out after I had send you my brilliant suggestion. Sorry.
Signature · 1 line
Markus
Quoted 6 lines · Frederik Himpe
I'm experiencing a lot of problems with audio skipping on my system. I am using Rhythmbox as an audio player and have Compiz active. Whenever I maximise or restore a maximised real transparent gnome-terminal, audio skips. Compared to other normal applications, restoring the transparent console is also slow (it takes a few seconds before the window appears).
Quoted 7 lines
I ran
cfs-debug-info-clear.sh, and the reproduced the problem 3 times by restoring and minimising a gnome-terminal window. After that I ran cfs-debug-info.sh, and I uploaded the information here:
http://artipc10.vub.ac.be/cfs-debug-info-2008.04.12-18.00.20.gzThis problem happens with both 2.6.24.4 as with 2.6.25-rc9 (the debugging information is from the latter version).
I once learned ( http://lkml.org/lkml/2008/2/29/179 ) that the CFS would have something to do with the latencies I was observing in my system if se.wait_max was bigger than 40 msecs. It would be an "anomaly" in CFS.
But in the log you uploaded you have only two se.wait_max greater than 40 msecs
[mafra@localhost:~]$ grep se.wait_max cfs-debug-info-2008.04.12-18.00.20 |awk '{ {if ($3 > 40.000000) print $3} }'
44.153039
50.841370As these numbers are only a bit greater than 40, I guess your problem is not related to the CFS scheduler.
Can you try latencytop? ( www.latencytop.org )
Bye,
Carlos
Quoted 11 lines · Carlos R. Mafra
But in the log you uploaded you have only two se.wait_max greater than 40 msecs
[mafra@localhost:~]$ grep se.wait_max cfs-debug-info-2008.04.12-18.00.20 |awk '{ {if ($3 > 40.000000) print $3} }' 44.153039 50.841370As these numbers are only a bit greater than 40, I guess your problem is not related to the CFS scheduler.
Can you try latencytop? ( www.latencytop.org )
I'm not too sure how I should use latencytop and interpret its results. I start latencytop, and reproduce the problem while rhythmbox is playing. It seems that it automatically changes the selected application in the bottom list after a few seconds, probably based on the CPU usage, where it selects most of the time either compiz or X. The upper half of latencytop is always empty, for all applications. In the lower half, it shows for all applications something like:
Process X (7972)
Scheduler: waiting for cpu 43.5 msec 43.5 msec
Only the times differ between the different processes.
And that's all...
Signature · 1 line
Frederik Himpe
Quoted 22 lines · Frederik Himpe
Quoted 11 lines · Carlos R. Mafra
But in the log you uploaded you have only two se.wait_max greater than 40 msecs
[mafra@localhost:~]$ grep se.wait_max cfs-debug-info-2008.04.12-18.00.20 |awk '{ {if ($3 > 40.000000) print $3} }' 44.153039 50.841370As these numbers are only a bit greater than 40, I guess your problem is not related to the CFS scheduler.
Can you try latencytop? ( www.latencytop.org )
I'm not too sure how I should use latencytop and interpret its results. I start latencytop, and reproduce the problem while rhythmbox is playing. It seems that it automatically changes the selected application in the bottom list after a few seconds, probably based on the CPU usage, where it selects most of the time either compiz or X. The upper half of latencytop is always empty, for all applications. In the lower half, it shows for all applications something like:
Quoted 4 lines
Process X (7972)
Scheduler: waiting for cpu 43.5 msec 43.5 msecOnly the times differ between the different processes.
Did you try with the git version of latencytop?
I remember that once I downloaded latencytop from the website and the upper half of latencytop was also always empty, and that the git version was better and showed the latencies.
Quoted 8 lines · Frederik Himpe
I'm not too sure how I should use latencytop and interpret its results. I start latencytop, and reproduce the problem while rhythmbox is playing. It seems that it automatically changes the selected application in the bottom list after a few seconds, probably based on the CPU usage, where it selects most of the time either compiz or X. The upper half of latencytop is always empty, for all applications. In the lower half, it shows for all applications something like:
Quoted 4 lines
Process X (7972)
Scheduler: waiting for cpu 43.5 msec 43.5 msecOnly the times differ between the different processes.
Had the same problem until just a few days ago when I figured it out mostly by accident.
You need to 'echo -n 1 > /proc/sys/kernel/latencytop' to enable latencytop tracing to get more useful info. It's something that's badly documented unfortunately. Make sure you disable it again when you're done so you don't keep the tracing overhead.
Not sure why starting the latencytop application does not enable tracing automatically, or at least provides an option to do so.
I also agree that the automatic switches to other apps are strange. Adding a particular app name or PID on the command line does not seem to have any effect.
Unfortunately the latencytop website also misses usage info and seems generally outdated as it still says that a patch to the kernel is required.
CC'ing Arjen in the hope that he can find some time to improve this.
Cheers,
FJP