[LAU] Realtime Kernel

[Florian Schmidt]

On Tuesday 18 March 2008, Arnold Krille wrote:
> Am Dienstag, 18. März 2008 schrieb Mark Knecht:
> > On Tue, Mar 18, 2008 at 11:25 AM, Arda Eden <ardaeden at gmail.com> wrote:
> > > So is it really necessary to have a realtime kernel in order to use
> > > linux audio applications without any latencies or xruns ?
> >
> > It is not possible to use ANY kernel with NO latencies. EVERY kernel
> > has latencies. The question is how low do you want the latencies to
> > be? If you can exist with 50mS or higher you might get away with a
> > standard kernel. If you want to run with 1.2mS latency then you will
> > absolutely have to have a real-time enabled kernel.
>
> Still it has to be noted that you can achieve 5ms latency without a
> RT-kernel... Even with an el-cheapo builtin soundcard...


The question is not really whether it works at all, but rather how reliable it 
is. Even if it is possible to run a vanilla kernel with a < 10 ms latency 
setting it will be far more reliable to use a properly tuned -rt system. With 
the -rt system it is pretty much guaranteed that the kernel nor any other 
processes might ever make jack fail to meet its deadline. With a vanilla 
kernel you might just have been lucky.

So even if i were to record with a high latency setting (e.g. i don't need low 
latency for monitoring or live instrument effect processing) i would still 
use a properly tuned -rt system. A single xrun can ruin an otherwise magic 
take/recording, a maybe unrepeatable performance. To minimize the chances of 
this happening is IMHO  almost always justifying the use of an -rt kernel 
when doing audio recording/processing..

Regards,
Flo


-- 
Palimm Palimm!
http://tapas.affenbande.org