Re: Sound server observations

[Stefan Westerfeld <stefan space twc de>]

    Hi!

On Thu, May 17, 2001 at 08:22:37PM +0100, Michel Alexandre Salim wrote:
> EsounD : low performance hit (~ 3%). Requires programs to support
> outputting sound to it
> aRtsD    : higher demand on CPU. Sometimes forces itself on other
> programs if not suspended first
> 
> aviplay totally unusable on my P2-333 if aRtsD is enabled, but weirdly
> XMMS can bypass aRtsD if it's output is set to OSS.
> 
> CPU usage for XMMS: 
>     running using OSS: < 1%
>     running under esd:  < 5%
>     running using aRts: > 10% (max. buffer) - 15% (standard settings)

Well, I guess your benchmark is wrong. Using "top" or similar to benchmark
threaded programs under linux doesn't work well (the cpu usage you see is
simply wrong), so... don't. ;) I did a similar benchmark with mpg123 and
the current artsd CVS version (Athlon 700):

mpg123 foo.mp3          (OSS):  ~98% idle
artsdsp mpg123 foo.mp3 (aRts):  ~96% idle  (buffers: 12 fragments w/ 4096 bytes)

Also note that artsd is slower if it needs to resample.

There is probably still room for optimizations in artsd. But in the end, of
course you will need to spend some cpu cycles on producing sound in app A,
transporting it to artsd, and processing it there, and you might always find
a low-end configuration where don't want to afford this (i.e. small linux
handheld). Although by the time GNOME2 is widely used, the standard machine
for a desktop user will probably be quite powerful.

> If this is a fairly typical experiment, can GNOME be modified perhaps to
> support running *without* sound server? Since GNOME 2 will probably have
> aRts support, adding another option should not be too difficult.

Using CSL (see www.arts-project.org for release announcement) applications
should run fine with or without artsd, so use this and be happy ;).

  Cu... Stefan
-- 
  -* Stefan Westerfeld, stefan space twc de (PGP!), Hamburg/Germany
     KDE Developer, project infos at http://space.twc.de/~stefan/kde *-