On 11 Dec 2001, Hans Meine wrote:

> aRts uses more than just a different "nice value" - in fact it tries
> to use no less than what standard OS provide.
>
> For what "realtime priority" means here, look at
> "man sched_setscheduler".

I stand corrected!  (well, kinda.. cause if that's linus's idea of
"realtime" I want my Amiga back!! :-))

I did note that ALSA has this rtc hack which binds the rtc interrupt to
itself.  I don't read enough of the kernel mailing list to understand why
this still has to be a patch or if there's any drive to make it standard,
but it seems happy on my box, not that I noticed a difference!

I mean, I don't mind if my apt install takes an extra few milliseconds to
run but when XMMS starts sounding like someone's beating my computer with
a stick it gets annoying!  Or worse still - if I hit "fire" in a game and
a bullet appears but the 'bang!' sound occurs 1/4 of a second later.
buffers suck :-(

It could be that XMMS needs to run as SCHED_FIFO too.  I'm guessing that
the reason it chops up with artsd but not when talking directly to
/dev/dsp is to do with priority of different types of message (again, I'm
not an expert on the kernel :-)) and the fact that artsd means there's two
data streams instead of one.

I suppose the best remedy for my problem is to try and persuade XMMS to
talk to alsa0.9 in O_NONBLOCK mode alongside artsd.  There's a 0.9 plugin
already out but it doesn't work on my debian box and I haven't got round
to hacking it yet.


Reply via email to