Re: OpenSound - was Re: lockmgr patch

["Dmitri Nikulin" <dnikulin@xxxxxxxxx>]

On 6/17/07, Dave Hayes <dave@jetcafe.org> wrote:
> Back in the day, we keyboard players could easily have 5-6 pieces of
> outboard synth rack gear in use at once. It would be really nice to do
> this virtually, someday, and muliple-core CPUs are most likely the
> low-level answer. DFly's project goals are great infrastructure
> support for this kind of application.

DragonFly's basic attitude towards kernel-side processing is
"aggregate work more". Which means the focus is on getting larger
chunks of work done at a time, so that the best use of cache and
reduced switching can be made, which is basically the opposite of
preemption. So unless a special architecture is put in place to give
audio recording a sort of preemptive capacity, it's doubtful that
DragonFly can make even pseudo-real-time guarantees.

But if Matt is right and all it takes is giving the audio handling
higher priority so the desync is minimised, then it doesn't seem like
such a big deal. But this does not match up with the rest of the
world's observation that it's a hard problem. So It may be that
DragonFly and probably NetBSD will be left out of the real-time space
for a while yet. I think FreeBSD is already well on the way to the
kernel side of pseudo-real-time sound but if it needs an ALSA-like API
then that could take some work.

---
Dmitri Nikulin

Centre for Synchrotron Science
Monash University
Victoria 3800, Australia