20 Oct
2004
20 Oct
'04
11:52 a.m.
Thanks for the reply, Mark.
I ran Benno's disk latency tests quite a bit when I was setting up my system. As far
as I could tell, the test just checks playback, not record. As I mentioned, I have no
trouble playing back (in either Duplex or Playback modes) to either disk system. When I
was fussing with the early 2.6 kernels, performance with these tests was always worse than
with 2.4.23. With the 2.4.23 kernel, tests on the IDE/ATA drive were consistently <
3ms, whereas there were usually a couple >3ms blips with the SCSI Raid0 with the
diskwrite and diskcopy stress tests (all other tests were <3ms). When I turned on the
write cache on the SCSI drives, the results were like the IDE/ATA, but I got just as many
xruns with jackd and ardour. This was one of the observations that made me wonder whether
the problem was actually the SCSI controller (i.e., nothing I did to the SCSI system
affected the number of xruns).
I haven't tried those latency tests with the 2.6.9 that I have now, primarily because
they didn't match up with the "real world" performance that I am concerned
with (multitrack recording). This is why I'm hoping the ecasound tests will tell me
something.
As for desktop, I'm using openbox - I got fewer xruns with it than with gnome. Never
tried KDE, but posts to this list (including yours!) don't encourage me to.
Joel
It's been quite a while since I've heard of
anyone using it, but what
about running Benno's disk latency testing program on your raid drives
and see what happens?
Or does that no longer work with 2.6 kernels?
When I worked with the tool it gave me very reasonable numbers vs.
what I was seeing for xruns.
Also, I don't think I remember you mentioning what desktop environment
you are using, but even with my 2.6.9-rc2-mm4 type kernel I got lots
of xruns under KDE, while Gnome was much better and fluxbox was, for
me, the best. (measurably...)