[pulseaudio-discuss] Distribution kernels and glitch-free (Packagers, read this!)

[Mark Greenwood]

On Tuesday 24 February 2009 13:34:19 Sean McNamara wrote:
> Hi,
> 
> On Tue, Feb 24, 2009 at 7:20 AM, Jan Claeys <lists at janc.be> wrote:
> > Op maandag 23-02-2009 om 15:18 uur [tijdzone -0500], schreef Sean
> > McNamara:
> >> * Once upon a time (old releases, e.g. 7.04 and below?) there was the
> >> -lowlatency kernel flavor. This one went all the way and gave us
> >> CONFIG_HZ=1000 and a fully preemptible kernel (not only voluntary, but
> >> forced preemption). I'm not sure of the rationale for discontinuing
> >> this kernel flavor, but it would be silly to say it's only because of
> >> mirror disk space or something; their disk space consumption has gone
> >> way up since then.
> >
> > Maybe have a look at the -rt kernel?  :)
> 
> The -rt kernel is in universe, and is not kept in sync with the kernel
> revision of the -generic, -server and -virtual kernels. Here are my
> observations about -rt so far:
> 
> *Grabbing linux-rt out of the jaunty repositories gives me a 2.6.27.3
> kernel, whereas the kernels in the "main" repository, -generic and
> -server, are 2.6.28.x kernels. There is some neat stuff in 2.6.28 that

This is probably because realtime is very borked in 2.6.28. Some of Ingo Molnar's work has been added to the mainline kernel but the rest of it is incompatible with some of the stuff in 2.6.28. There isn't even an -rt patch for 2.6.28 on kernel.org (or wasn't, last time I looked).

I'm under the impression that most people doing any serious low-latency work (eg JACK users, that kind of thing) are mainly still using 2.6.24 because that's the last really solid rt flavour. I know I tried more recent versions but they don't hold up so well.

But anyway, going RT is somewhat overkill for the majority of people, I suspect. We don't all need 0.6ms audio latency.. ;)

The thing that concerns me most is distros still not using Hz_1000. Some audio apps (notably Rosegarden) complain like hell if you don't have Hz_1000 because their developers know the app won't get the kind of response it needs. The last experiment I did, with powertop, the difference in power usage between a 300Hz kernel and a 1000Hz kernel (all other things being equal) was about 100mW. Perhaps that's significant on a server running 24/7, but not for a desktop or laptop when using Hz_1000 appreciably improves the responsiveness.

Personally I don't mind being told to build my own kernel because that's normally what I do anyway... but it's not appropriate for most people.

Just my thoughts.

Mark