[ardour-dev] Latency: AGP vs PCI video card and XFS journalling vs not

[Tommi Sakari Uimonen]

> beta9+3) with a 2.6.3 kernel and have played around w/ all the

Try 2.4 series with lowlatency & preemtive patches. AIUI, 2.4+ll+pe is
still better than any 2.6.

> standard latency issues (no extra stuff running, swapped PCI cards
> around for interrupts, etc.).  However, I'm not able to go lower than

irq 9 would be best for soundcard. The irq order for non-apic system is,
from highest priority to lowest: 1,8,9,10,11,12,13,14,15,3,4,5,6,7

> 2.  Could it be my XFS filesystems?  If I have 8 or so tracks on one
> drive, it can't even handle playback w/o giving disk errors.  I've
> tweaked all the XFS options, and currently have a large (64MB) log for
> each filesystem that resides on the opposite disk.  Would turning the
> log size up or down help?  Is there a way to just turn off journalling
> in XFS, or would ext2 be better?  I did notice that exporting to my
> default param (i.e. small log) ext3 root partition seems to be faster
> than to my XFS.

For recording purposes, "write as much as possible, as fast as possible",
(so no reading other tracks at the same time, just writing) I found ext2
to be fastest (ext3,reiserfs were included in the test), since it doesn't
spend time journaling.

Actually I tested this with my MiniDV, which gives 25MBits/s through
firewire, which corresponds roughly to 11 channels of 24/96.

So this test was only about disk performance, not JACK+Soundcard involved.

But as a conclusion from the results, I used ecasound with 10 channels of
24/96 (stored the data as 32bit) to ext2 and never got any
problems.

I don't know if ext2 is good for simultaneous r/w, nor ext3 and reiserfs
for this matter.

Tommi Uimonen