[LAD] [CFP] Linux Plumbers Conference CFP ends soon!

Lennart Poettering mzynq at 0pointer.de
Sun Jun 14 22:53:54 UTC 2009


On Mon, 15.06.09 00:01, Fons Adriaensen (fons at kokkinizita.net) wrote:

> 
> On Sat, Jun 13, 2009 at 12:16:22PM +0200, Jörn Nettingsmeier wrote:
> 
> > that said, lennart, any chance you could make it to a LAC anytime soon?
> > pa has got some heat lately, some deserved imho, some not, but it
> > definitely serves a very palpable demand.
> > i believe that after some time it can be a real blessing for casual
> > users without necessary having to become a PITA for pro audio users -
> > but for that, it would be nice to discuss the mutual needs and wishes
> > over pizza. [2]
> 
> I agree. I'm not a PA user - I dislike 'desktop sounds' and any
> serious app I expect to use Jack. But if PA could provide an
> interface that would allow Jack (or other apps) to connect to it
> *without any loss of performamce* compared to direct access of
> an Alsa hw: device, [1] then I wouldn't mind having it installed
> and being used.

The part about "without any loss of perfomance" is the big problem. Of
course the latency increases each time you add an indirection. There
is no way around this. A corollary of that is that PA and JACK need to
have the same "distance" from the low-level device. The only way I see
how that could work in an intregrated solution is if JACK and PA would
merge in some way. While that might be a nice idea (which I certainly
don't oppose) I don't see that coming in any way. It's simply that the
playback models of PA and JACK are simply too different. PA uses very
large hw buffers (2s), rewinding, timer-based audio scheduling (which
allows us to adjust the wake up intervals dynamically to what the
clients request), zero-copy IO, non-float sample types, and so
on. That's all very useful for normal desktop use and for minimizing
wakeups/power usage/CPU load. OTOH JACK's primary objective is low
latencies, and power usage or non-floats, ... don't matter at
all. Which effectively means sound card IRQ based scheduling and short
buffers make much more sense. A hybrid engine that could do all of
this and switch between timer-based scheduling and sound card IRQ
based scheduling in one would probably be exceptionally complex, and I
see noone who would be willing and have the time to work on that. [1]

The only other way I see to give PA and JACK the same "distance" from
the device is if PA goes out of the way when JACK runs. Which is
basically what JACK2 and PA do now. It's a however only a pragmatic
compromise, not more.

Not many people work full-time on audio for Linux. Given that this is
how it is pragmatic comprimises are probably what we have to work
with.

>  Pizza ? In Holland ? More likely a 'broodje kroket'.

So I read from that that LAC 2010 happens in .nl?

Lennart

[1] Canonical appears to be looking for someone to hack on PA full
    time. So if someone wants to attack this issue, he might
    even get some funding for it if he does so under the PA umbrella.

    http://webapps.ubuntu.com/employment/canonical_DASE/

-- 
Lennart Poettering                        Red Hat, Inc.
lennart [at] poettering [dot] net
http://0pointer.net/lennart/           GnuPG 0x1A015CC4



More information about the Linux-audio-dev mailing list