irq priority for firewire device
ronan at jouchet.fr
Sun Oct 24 04:49:46 BST 2010
On 10/22/2010 06:44 PM, Gerhard Lang wrote:
> would it be possible to make something like "rtirq" working in a full
> preempt and noticks-disabled vanilla kernel without rt-patching?
> As far as I can see adjusting IRQ priorities seems to be essential for
> firewire-audio as long as firewire's IRQ-priority is subordinated at the
> low end by default, and if this could be changed in a user friendly way
> I'd see no more need for rt-patching kernels.
> I compared all kernel mods for audio live performance and recording in
> practice, and the only feature I really miss in non-rt-kernels is this
> rtirq thing. But I'm no coder and need guiding hands for understanding
> deeper kernel and os matters, so please bear with me if this idea should
> be foolish, misplaced or redundant/obsolete.
> Best regards Gerhard
Mentioning benchmarks of kernel mods rings a bell here. If you kept some
numbers, please share them with us, I'm certain many people here would
love to read your findings.
Quick summary to help you catch on the recent discussion: UbuntuStudio
is currently without realtime kernel because -rt is a pain to maintain.
Hopes and efforts are crystallizing around -lowlatency (which is
actually a normal -generic kernel with some compile-time configuration
tweaks but none of Ingo's IRQ kernel magic). However, many people don't
know -lowlatency, fear losing performance compared to good ol' -rt, and
see this change as an obvious loss. Complaints / lack of real test data
/ sterile debate ensues. Nevertheless, the few people having tested this
new kernel (including me) reported surprisingly good results.
Which leads me to my question about your message: you say that according
to your tests, the rt-patch (i.e. adjusting irq priorities) is
*essential* for firewire audio. Like what? How much latency did you win
with these kernels compared to recent -generic or -lowlatency kernels?
How many xruns avoided? On which hardware?
Thanks for your participation, we'll really appreciate additional kernel
benchmarks. Finally, maybe abogani will be able to answer your
kernel-related questions; try to catch him on #freenode if he doesn't
see this message.
-- Ronan Jouchet
More information about the Ubuntu-Studio-devel