Even if something else is causing my overruns now, I
certainly suspect
that
real-time performance could be enhanced by a bit of
cache settings
tweaking.
If anyone has any info on doing this with a 2.6.test5
kernel, I would love
to get it. My curiosity has been awakened :)
I ought to be in bed right now, but I did manage to find what I was looking
for. The values I want to adjust are in /proc/sys/vm/. Some basic
observations about the effects of different cache settings on I/O
performance can be found at:
http://drscriptt.riverviewtech.net/linux/enotes/bdflush.html
I also found a page with some other potentially useful information for
tracking down hard disk I/O operations (although the goal of the document is
quite unrelated to audio streaming):
http://www-2.cs.cmu.edu/~mukesh/hacks/spindown/t1.html
Lastly, I changed the IO_support setting to 1, and Ardour has now been
happily recording for an hour and a half! I was also able to run the
latency test without overruns (the 4 ms. spike dropped to 2.7 ms.). Thanks
so much for your help, Mark & Jan!
I'll be looking into the disk caching parameters, and if I come up with any
promising results, I'll certainly post my findings.
Best regards,
|)
|)enji
Benjamin Flaming
--------------------
"The trouble with computers, of course, is that they're very sophisticated
idiots."