Subject: Re: [PD-announce] Re: [linux-audio-dev] ANN: k_jack v0.0.0.5 and Mammut v0.15
From: Kai Vehmanen (kai.vehmanen@wakkanet.fi)
Date: Thu Jan 23 2003 - 23:08:28 EET
On Wed, 22 Jan 2003, Kjetil S. Matheussen wrote:
> For two years (or somethings), people have complained about the bad
> performance of the jack system. And I don't think it has been solved. I
> dont know about alsa; and doesn't understand the driver-code, so it was
> easier for me just to reimplement jack.
First of all, I must say I have nothing against k_jack. I don't think I
fully understand the motives (or the mechanims or design features that
enable better performance), but as Ron Kuper wrote today, competition is
good. I don't think it's fair to expect commercial companies to deal with
competition from FOSS groups, while not welcoming competition from other
FOSS developers. ;)
And yes, there are performance problems with JACK in certain use
scenarios:
1. Running with root privileges (and -R) but with
a standard kernel (no low-latency or preemption patches).
In RT-mode jackd doesn't tolerate client failures. But
if you don't have a low-latency kernel, you will get
50-100ms stalls during normal use. In other words,
running jackd with period size less than 50-100ms
(-p 2048 => 46ms at 44100Hz) _WILL_NOT_ work reliably.
Solution1 => Patch the kernel.
Solution2 => Increase period size (-p 1024 or higher).
Solution3 => Increase period count (-n 4 or higher) and
run without root privileges.
2. Running complex setups without root privileges (and -R).
If you have X clients participating in one JACK
setup, those X clients have to get the CPU during
each engine cycle. Without root privileges (and SCHED_FIFO
scheduling policy), there's no magic that makes sure
that all clients get enough CPU time. So the more
the processes you have, the larger the chance of
a client failure.
Solution1 => Increase the period size or period count.
These _HAVE_ to be notably higher than
with single-thread audio apps.
Solution2 => Run as root (see previous problem scenario).
Solution3 => Patch the kernel and run as root.
> k_jack performs okey. You dont have to run things with realtime priority.
What does "cat /proc/asound/card0/pcm0[cp]/sub0/[hs]w_params" output
while you are running k_jack? This information should help us understand
the benefits of k_jack over current jackd.
-- http://www.eca.cx Audio software for Linux!
This archive was generated by hypermail 2b28 : Thu Jan 23 2003 - 23:15:50 EET