[ARM] Pulse Audio eating up to much CPU

[Jerone Young]

On the Freescale MX51 platform it has been noticed that Pulse Audio eats up a signficant amount of CPU that is unacceptable. This is using the now offical images being produced for the Freescale platform here:
https://wiki.ubuntu.com/BabbageJauntyBetaInstall

I did some findings:

When playing the ogg theora video of big buck bunny. Pulse Audio eats up 15 - 30 % CPU.
http://mirrorblender.top-ix.org/peach/bigbuckbunny_movies/big_buck_bunny_480p_stereo.ogg

Pulse audio is eating only 1 - 2% CPU when playing ogg files such as
http://www.vorbis.com/music/The_Abyss-4T.ogg

Also playing mp3s within the software codecs in packages:
gstreamer0.10-ffmpeg
gstreamer0.10-plugins-ugly
When playing through totem pulse audio will then eat up
a bit more averaging 2 - 6 % CPU.

[Jerone Young (jerone)]

Freescale engineer is reporting pulse audio consumes 30%-40% CPU during video playback with latest build as of april 12 09.

They have now removed pulse audio using aptitude, and are no longer using it.

[Jerone Young (jerone)]

For ARM the reason this is important is that Pulse Audio eating significant CPU means less battery life for ARM based devices. So we need to minimize the CPU utilization of Pulse Audio on ARM.

[Daniel T Chen (crimsun)]

Pending mainloop and threading rework in upstream git HEAD (i.e., won't be
resolved in 0.9.1[45])

On Apr 13, 2009 4:05 PM, "Jerone Young" <email address hidden> wrote:

For ARM the reason this is important is that Pulse Audio eating
significant CPU means less battery life for ARM based devices. So we
need to minimize the CPU utilization of Pulse Audio on ARM.

-- [ARM] Pulse Audio eating up to much CPU
https://bugs.launchpad.net/bugs/358831 You received thi...

[Martin Pitt (pitti)]

Nobody in the desktop team has an ARM device for debugging this. David, can someone from your team attach strace/gdb to this to check where it is burning so much CPU power?

[Oliver Grawert (ogra)]

there are no audio drivers on the babbage1, here it defaults to /dev/null output, pulse doesnt show such a behavior. what kernel version was used for the above test ?

[Jerone Young (jerone)]

Testing was done with SoC vendor provided kernels. Both at version 2.6.26 & 2.6.28. Also same problems are being seen on babbage 2.

Since babbage 1 has now been end of lifed and is no longer used, nor will be shipped to any users or customers.

[Oliver Grawert (ogra)]

please make sure to test with the distro provided kernels, we can't do anything about misconfiguration in kernels we do not maintain. also let me point out again that the build-babbage-rootfs script is not a valid way of building images beyond a basic test state, there were various issues with audio handling in jaunty for quite a while due to missing group membership of the default user. the only images we can take bugs on are the officially released ones, babbage2 support will be added with the alpha3 milestone (according to the kernel team) in karmic, feel free to re-open the bug if the problem presists with one of the official images using the officially released kernel.

[Jerone Young (jerone)]

This is still very much a valid bug. We need to add audio support to our supported kernel. I have been in deep descussion with David Chen (of pulse audio fame) over this issue.

When we have audio support in our kernel and we can't see the problem, then we should close the bug.

Otherwise this is very much a valid problem we need to solve.

[Oliver Grawert (ogra)]

please leave the bug state until you can test against a valid audio stack, kernel and image we support