24 Jul
2011
24 Jul
'11
7:09 p.m.
On Sun, 2011-07-24 at 18:59 +0200, Robin Gareus wrote:
Hi Ralf,
On Jul 24, 2011, at 4:12 AM, Ralf Mardorf wrote:
Thank you for the explanation Robin :). Cheers! Ralf
3.0-rt2
So, the "rt patch emulation" ex kernel 2.6.39 might be less good than a
kernel that is patched?
It's not "emulation". A big part of the RT patch (e.g. threaded IRQ
handlers) was been merged into 2.6.39.
The goal of the Preempt-RT patch is to eventually end up completely in the vanilla (aka
mainline) kernel.
http://lwn.net/Articles/440064/ is a nice read on the current state of it.
Until now I build
linux-image-2.6.33.9-rt31_2.6.33.9-rt31-10.00.Custom_amd64.deb
linux-image-2.6.39.1_2.6.39.1-10.00.Custom_amd64.deb
but I only produced music with 2.6.33.9-rt31 until now, dunno if 2.6.39
isn't as good as a kernel-rt.
Any information, experiences?
I've been running 2.6.39 ever since it came out and used it for production. I
can't tell a difference between it and the rt-patched 2.6.33 for general audio/video
production. Neither will produce x-runs - unless I do sth stupid or overload the system.
Threaded IRQ handling of the audio-interface and [high priority] FIFO scheduling are the
important parts for low-latency audio and they're mainline in 2.6.39 (FIFO scheduling
was for a long time already).
There's a lot of small (but complex) things the RT patch does that are not in
mainline yet (e.g pre-emptible memory-management, sleeping spinlocks) and can improve the
guaranteed response-time of dedicated [audio] processes. However the gain [of code
available in RT vs what is already mainline] is rather small when compared to latency
requirements of audio (us vs. ms) and thus can be neglected on most systems. It'll
probably give you an edge though.
Long story short: If you just want to make music: stick with 2.6.39 or 3.0 - If you have
the time and ability to debug: give 3.0-rt3 a go.
Even though the impact of the RT-preemt patch on pro-audio is becoming less dramatic.
It's a great project and important project.
2c,
robin