[linux-audio-user] low latency/preemption

Florian Schmidt mista.tapas at gmx.net
Tue Jan 31 03:58:32 EST 2006

On Mon, 30 Jan 2006 16:49:34 -0800
Mark Knecht <markknecht at gmail.com> wrote:

> > As I read it, the low latency kernel patch(es) only apply to kernel 2.4,
> > and with kernel 2.6 the only option for
> > improving latency is kernel pre-emption.
> >
> > Have I got this much right?
> >
> > If so, is it better to use a patched 2.4 kernel than a 2.6 kernel with
> > pre-emption?
> >
> > TIA, Mick
> Hi Mick,
>    Which is better will probably get some discussion, but from a
> practical point of view I'd say that if your distro uses a 2.6 kernel
> then you should stick with 2.6. There are probably too many library
> issues and other things beyond my limited knowledge that will bite
> you.
>    I think that most people running a 2.6-ll type kernel will tell you
> it works pretty well most of the time. ;-)


the new deal is the "realtime preemption" patches by ingo molnar. These
are available only for 2.6.x kernels. If they run (and they run very
well here) on your box you can expect buffer sizes of 64 frames and
less(!) to work reliably. I know, some people don't believe me. but it
is possible for me to use a buffersize of 8(!) frames reliably with my
delta 66 card. Of course this isn't really useful as context switching
overhead gets kinda expensive (jack uses 20% cpu time when idle with
this buffer size), but it proves the superiority of realtime preemption
patched kernels over any 2.4.x kernel (even with ll and prremption
patches applied).

On 2.6.x you also don't have to go through the capabilities pain to get
non root users to be able to start realtime processes.

I'd recommend searching for "realtime preemption", "realtime lsm",
"rtlimits" to get more info (or have a look at my site in the signature
where i have collected some info).

Have fun,

Palimm Palimm!

More information about the Linux-audio-user mailing list