[linux-audio-user] The trouble with disks

[Robert Jonsson]

Did you put jacks tempfiles on a tmpfs partition ?
I regularily change /tmp to a tmpfs partition in /etc/fstab these days.

none /tmp tmpfs default 0 0

or something like that...

/Robert


torsdagen den 18 december 2003 11.35 skrev John Anderson:
> I'm using ardour to do around 8 tracks of audio, not all have regions
> active all the time. Playing a session through the first time, I get
> grrrtz every so often, usually but not always accompanied by an xrun
> report from jack. It's correlated with disk activity. Which is a Fujitsu
> MAN3367 on an Adaptec 29160 and according to hdparm can do around
> 50Mb/s. Although in ardour the disk throughput indicator shows anything
> from 9 to 14, depending on the number of regions being played, and the
> scheduler settings in use. I guess that's because the disk has to seek
> for the data for various different regions. Which are on a Reiserfs
> partition.
>
> I'm using an Athlon 2200+ on an MSI motherboard with 1Gb of memory. This
> is after my nice dual-athlon motherboard died, which had none of these
> problems. <sigh>
>
> I'm using 2.6.0-test11 kernel (with pre-emption), and I've tried the as
> scheduler (with various settings), the cfq scheduler and the deadline
> scheduler (which seemed to be the worst for this kind of low-latency
> work). It's subjectively better than 2.4.22 with low-latency and
> pre-emption patches.
>
> top shows that whenever there's a dropout, the CPU has just spent
> anything from 6 to 30% on io wait.
>
> It's definitely not a sound card issue - once the session has played
> through once and the audio files are cached, I get flawless playback
> with ardour showing disk throughput of 190Mb/s, even with a compile
> going in the background. I spent a few days playing with PCI latencies
> as well, but that didn't make a difference. Dropping the SCSI controller
> tagged queue depth to 4 makes recording more reliable, but doesn't
> really help otherwise.
>
> It also doesn't seem to make much of a difference if I run jack with
> buffer size of 1024 or 256. I haven't tried lower than that, and the
> card (Terratec EWS88MT) won't go higher that -n 2 -p 2048.
>
> Short of getting another dualie motherboard (which is a PITA in this
> country, and expensive), is there anything else I can try? low-latency
> patches for 2.6 kernels?
>
> thanks
> John