I had sound drop-outs when playing PCM data
from a large RAM buffer (not from disk!) , to a 24k sound buffer (3 x
8k fragments),
I have an interactive application which must have low-latency (for
example
user can change the playback speed in relatime and so on .)
The ram-player is scheduled with realtime priority s normally there are
no drop-outs,
but since I must rip audiotracks from a Plextor 20x SCSI CDROM in
background,
there is quite heavy disk write acrivity. (the disk is a IBM 6.4GB ATA
UDMA)
(I use Linux 2.2.4 the SCSI controller is an AHA2940)
My problem was : everytime cdda2wav finish to rip a track, the file
buffers gets synced, and
the sound drops out for a fraction of second, due to the fact that
kernel blocks the for too much time,
and blocks even real-time scheduled prcesses.
I tried to play with some programs to improve audio performance:
1) hdparm -u 1 /dev/hda : sound dropouts still present
2) /sbin/update -f 1 : sound drop outs still present
3) inserted an fsync / fflush after each buffer write in cdda2wav :
EVEN WORSE , still many sound drop-outs
4) launched this simple script in background
while true ; do
sync
sleep 2
done
AND THIS IS THE ONLY THAT WORKS !!
I monitor constantly (testing this with OSS) the struct
audio_buf_info.fragments
which gives me the used fragments,
Normally this value =0 , using the above script to sync every 2 second,
there are no more audible sound drop-outs but the fragments value goes
to 1 or 2
when cdda2wav finish to rip a track.
Do you know if there are some plans for future kernels,
to avoid blocking realtime scheduled processes for more than 10-20ms
during heavy disk writes ?
I think this is a feature which is quite often needed in real-time
multimedia applications,
especially harddisk recording in conjunction with playing low-latency
audio.
take for example an audio sequencer alla "Cubase VST" and a realtime
synthesizer like "Generator".
with enough horse-power , under windoze you can run these 2 applications
at the same time and
record audio while "Generator" plays in realtime (emulating an analog
synth) the notes you play
on a midi keyboard (with low-latency 50ms)
I think Linux should provide a base, to allow running such applications
smoothly at the same time,
so that we have a rock-solid platform to run not only web/file servers
but high-performance audio
applications.
Do you know if the kernel developers have plans to address this issue.
thank you for infos,
regards,
Benno.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/