[pulseaudio-discuss] Distribution kernels and glitch-free (Packagers, read this!)
lennart at poettering.net
Sun Feb 22 12:24:45 PST 2009
On Sun, 22.02.09 20:01, Mark Greenwood (fatgerman at ntlworld.com) wrote:
> On Sunday 22 February 2009 19:37:11 Lennart Poettering wrote:
> > Heya!
> > As one result of the alsa-time-test testing (see that last mail of
> > mine regarding broken sound drivers) input I got from folks, I learned
> > how very different the different distribution kernels actually
> > behave. They are much more different than i actually assumed.
> Very, very, interesting.
> As someone who has done a fair amount of audio work on Linux I
> normally build my own kernel as a matter of course, since none of
> them (even the ones that claim to be 'real time') are really up to
> the sort of low latency stuff I'm trying to do. (That said, I've
> never tried Fedora :) ). I've found it quite possible to achieve
> latencies of < 1ms on decent hardware, so yes, 210ms is pathetic :)
> On the other hand, that system uses nvidia drivers so I think
> they're less of an issue.
> Lennart, are there any definite things that you feel should be built
> into a kernel, or any things that definitely should not be in a
> kernel, for g-f to work properly? I have a short list of things I've
> decided must not be defined in my RT kernels for audio work. I'd be
> very interested in doing some experiments with the Mandriva and
> Ubuntu setups I'm using because g-f doesn't seem to be working very
> well on either of them.
Sorry, I am not that much of a kernel developer, so I cannot reliably
tell you which options are really a must and whiches ones are
not. However, there are some obvious things: HZ=1000 and preempt.
Without too much fiddling it should be easy to get a stock kernel with
the appropriate config options that does 5ms of latency for the usual
cases just fine.
Lennart Poettering Red Hat, Inc.
lennart [at] poettering [dot] net ICQ# 11060553
http://0pointer.net/lennart/ GnuPG 0x1A015CC4
More information about the pulseaudio-discuss