rt kernel

[Ralf Mardorf]

Hi all :)

On Mon, 2011-04-04 at 14:51 +0200, Alessio Igor Bogani wrote:
> Hi,
> 
> 2011/4/3 Gerhard Lang <lang.gerhard at gmail.com>:
> [...]
> > buffersize/xrun ratio. For tedious audio work on standard hardware I see no
> > more need for -rt kernels.
> 
> I would be very happy if we could live without -rt kernel.

Reading the mailing lists I've got the impression that we still do need
the RT patch. I experienced even the patched kernel as not RT capable.

> > I guess there might be a little profit for audio
> > performance by compiling actual kernel cgroups disabled, no tics disabled,
> > preemptible kernel enabled and  timer frequency set to 1000hz.
> 
> This is exactly the -lowlatency kernel.

Note that today we should prefer the tickless timer. I'm enabling it on
Maverick by my session start scripts:

lsmod | grep snd_hrtimer
 case $? in
    1)
	echo "Enabling HR TIMER for ALSA MIDI playback queue timer resolution
1000000000 Hz"
	sudo chgrp audio /dev/hpet
        sudo chmod g+rw /dev/hpet
	sudo modprobe snd-hrtimer
	;;
    *)
        echo "HR TIMER for ALSA MIDI playback queue timer resolution
1000000000 Hz is enabled"
        ;;
 esac

The 1000 Hz timer still isn't obsolet, because HPET/HR timer sometimes
does cause issues on some machines, resp. when running some apps, e.g.
Rosegarden will freeze the computer, but e.g. Qtractor is working.
Anyway, using the tickless timer when ever possible, will cause
something that is nearer to 'hard real time'.
HPET is enabled for Maverick's kernel 2.6.35-28-generic too.

FWIW compiling 2.6.33.7.2-rt30 for Maverick is easy to do, I posted a
scrip on that list.

What is my point?

RT on Linux isn't 'hard RT', so de facto even the RT patch isn't a real
RT patch.

> 
> > optional manipulating devices' IRQ assignment and priorities like this was
> > done by rtirq should be a feature for standard kernels.
> 
> I hope to provide soon a new -lowlatency kernel which provide that
> feature (which is available in 2.6.39).
> 
> Ciao,
> Alessio

By theory it's possible to edit an order, not only for 'the sound card
is more important than USB' etc., but also for e.g. 'make USB slot 3
head of the USB slots'. I'm using rtirq, but I'm not sure, that it does
have noticeable effect. IIRC sometime ago somebody wrote that rtirq only
is noticeable when using firewire, but I might be wrong.

OT: At the moment I've got issues with a Linux vocoder. I get xruns :(.
I experienced JACK or JACK clients as being the cause of most serious
trouble, but not that often the kernel.

2 Cents,

Ralf