Re: [linux-audio-dev] Re: Low Latency Kernel Combos

New Message Reply About this list Date view Thread view Subject view Author view Other groups

Subject: Re: [linux-audio-dev] Re: Low Latency Kernel Combos
From: Benno Senoner (sbenno_AT_gardena.net)
Date: Fri Jun 01 2001 - 17:56:39 EEST


On Fri, 01 Jun 2001, you wrote:
> hello,
>
> im curious about this low latency stuff.
>
> im running csound and timidity together, and there are times when the
> sound hacks badly, even though the cpu isnt very loaded.
>
> is this kind of thing solved with this low latency patch?
> (given that i run both in realtime mode?)
>

The lowlatency patch helps but does not guarantee that the apps will be
glitch-free, even if you run both apps in realtime mode.
The apps must be designed to take advantage of low latency and unfortunately
most of the existing ones are not.
The audio thread needs to perform only deterministic operations while leaving
out non-deterministic stuff like disk i/o , dangerous syscalls (which might
wait or cause a reschedule ecc).

This is the whole reason why we are trying to come up with some kind of
framework that is both flexible and guarantees dropout-free low latency audio.
Unfortunately the holy grail has not yet been found.

But the experience with my lowlatency testing stuff tells me that it will be
almost impossible to get 3msec latencies in an enviroment where multiple
processere are involved.
This is why I'm supporting the audio server concept where applications run like
plugins. This way we can be sure to get the latency we need but at the expense
of making the programming of realtime audio applications (=plugins) a bit more
complicated. (but not THAT more complicated).

Along with David Olofson we talked about using the audio server concept on an
SMP box or a cluster and as long as we run only one thread per CPU (or per
machine) and we have good algorithms that maximize parallelism, we can get
almost the same latencies as on an UP box.
The latter is true for an SMP box, while clusters depend heavily on the speed
of the networking interface. (but with 100Mbit/1Gbit nics you can do pretty
neat stuff cutting latencies down in the single digit range)

Sorry for going a bit of topic ..
:-)

cheers,
Benno.


New Message Reply About this list Date view Thread view Subject view Author view Other groups

This archive was generated by hypermail 2b28 : Fri Jun 01 2001 - 18:45:12 EEST