On Wed, Apr 09, 2008 at 02:40:04PM -0700, Fernando Lopez-Lezcano wrote:
> All this said, I personally haven't hit any issues with
things like MIDI in
> Fedora where the -rt kernel would have helped me. Maybe my gear is special,
> but latency between me hitting a key on a synth and having that note show
> up in rosegarden, and a sound being created is well below perceivable.
And that would be how many milliseconds?
I honestly couldn't say (it's been some time, and I'm a few thousand miles
from home right now).
What could seem below the perception level to you (caveat: I
don't know
if you are a musician and what instrument you play) may bother a
professional percussionist using the computer as an instrument. It all
depends on your demands as a performer.
"musician" may be stretching my abilities somewhat :)
But yes, I understand that I'm probably not pushing things perhaps as
far as some others may be.
Besides realtime performance, all the audio-over-the-network being
done
at CCRMA needs it (see:
http://ccrma.stanford.edu/groups/soundwire/).
This is, in a nutshell, multichannel (between 4 and 8 channels)
non-compressed high quality bidirectional audio being sent between 2 or
3 geographically separate locations to create a virtual concert hall and
jam session or concert. You don't want to add _any_ latency to the one
that speed of light already gives us :-) We routinely run at 64 frames
(or 128 if the links are not good). The stock Fedora kernel is just not
good enough although I'm sure it is getting better all the time.
*nod*. A while ago, I was tempted to add the latency-tracer part of -rt
to Fedora kernels just to see what would show up, but Ingo wasn't really
interested because he felt that we'd get a lot of reports we'd already
fixed in -rt.
The fundamental changes to spinlocks is probably the real controversial
stuff that's left to go upstream. How well that goes remains to be seemn
over the coming months.
> Right now the guys working on that stuff typically have a
> bunch of 'boring' test cases more tailored towards replicating situations
> like stock trades and the like. If we can construct additional use-cases
> I'm sure Ingo, Thomas & co would be very interested to hear about them.
> Especially if these cases are triggering different latency paths.
The apps being worked on at CCRMA (jacktrip) would touch both the
network drivers and drivers for pro audio cards. Probably disk as well
if the same host is being used to record the performance in Ardour.
Ok, that's an interesting use-case and in honesty, I don't think it's
too different to the use-cases like the stock trading stuff that people
have been focusing on, so it's understandable why you're reaping
the same rewards.
Dave
--
http://www.codemonkey.org.uk