[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: realtime kernel for Debian



Giacomo A. Catenazzi wrote:
Raphael Hertzog wrote:
On Tue, 24 Mar 2009, Giacomo A. Catenazzi wrote:
Do you really need real time kernel?
Debian is a technical driven project, but reading the previous two quotes,
"real time" is used as marketing thing.

It's good to question the use of any feature, but a real-time kernel is
certainly very useful in many industrial applications and Debian is
popular in that field. (Don't put a marketing label on anything where
you are not yourself sure of your expertise.)

Yes, I didn't write very well my sentence: the previous quotes was more
about "there exist rt kernels", "ubuntu has a rt kernel", but not solid
requirements. I had to write some "seems", and I'm sorry for the two
quoted people if it seems an attack.
Anyway, later in the mail, I asked for precise needs, so we could see
better what we should improve.

IMHO most users want a low latency kernel, but not a slower kernel, so
a CONFIG_HZ_1000 would be nice.  But the original post was about
multimedia production (and not reproduction), so the needs are probably
other.

My point was more:
- Debian has not rt kernel. Why? Non DD interested or/and low demand?
  This is an important point. We must not produce a rt-kernel if
  we cannot provide testers and developers (in unstable).
- kernel management is a weak point in distribution: no good method
  for kernel dependencies, using full capabilities, ...

so IMHO we should try harder with the normal kernel, so that we
can use the same infrastructure and testers. If we fail and we
are able to support rt kernels, IMO it is good to provide it in Debian.

The original mail was about "multimedia production" and few year ago kernel
developers had a lot of interaction with music industries.
I'm not an expert in the field, but how far are we in their need with
standard kernels?)


I do use a real-time kernel on a Debian based system for one of my
customers (but I have to recompile the kernel anyway because I do other
customizations) and I have good reasons to do so because I can't suffer
serial overrun and I must ensure that the serial interrupt handler
is run in the required time and that no other (kernel) task has higher
priority.

These *other customizations* are important to rt-kernel. So we need
a person (or more) that know the needs and could support us.
"realtime" alone is only a label ;-)


Thanks for the reactions so far! I think the newer kernels are improved for realtime (for audio usage, real low latency etc.). And there was some discussion about better realtime support in default kernels:

http://linuxmusicians.com/viewtopic.php?f=24&t=490

I think 95% of the users of the linuxaudio.org community (LAU mailinglist) uses a realtime kernel (CONFIG_HZ_1000 + Mingo patch (!?)). Discussion if it is still needed bumps up there once in a while, for example:

http://linuxaudio.org/mailarchive/lau/2009/3/10/153190

But till now people reports better results (mostly in terms of latency and xruns for jackd) with a patched kernel.

I know two people has started working again on rt patches for the newer kernel:

http://lwn.net/Articles/319544/
quote:
The realtime preemption project is a longstanding effort to provide deterministic response times in a general-purpose kernel. Much code resulting from this work has been merged into the mainline kernel over the last few years, and a number of vendors are shipping commercial products based upon it. But, for the last year or so, progress toward getting the rest of the realtime work into the mainline has slowed. On February 11, realtime developers Thomas Gleixner and Ingo Molnar resurfaced with the announcement of a new realtime preemption tree and a newly reinvigorated development effort.

Merging into the mainline kernel would be the best imho. So people who want to record stuff, realtime with fx, can use the default kernel. It would make a lot of people pretty happy.

Kind regards,

\r


Reply to: