[PlanetCCRMA] Roadmap to 2.6 kernel?

[Fernando Pablo Lopez-Lezcano]

> > Nando, could you elaborate a bit on the pros and cons for going to 2.6, please?
> 
> You mean for Planet CCRMA? Right now? Hard to say as I have not used 2.6
> extensively (just one build I did a while ago). From what I have heard
> it is not yet to the level of latency that 2.4+lowlat+preempt is at
> today, although it is very usable (again, not personal experience).
> Comments seem to indicate is "feels" faster than 2.4.x, and that's a
> good sign. 
> 
> I'd love to do a test 2.6 kernel... I'be been meaning to take a look at
> what's involved for a while... 

And I took a look. 

I briefly tested 2.6.4 + 2.6.5-rc2 under FC1 with the built in ALSA and
with the LSM module added to the kernel build (to run jack with realtime
priority as a non-root user). The latency performance is indeed worse.
No hard numbers, just a brief test on my laptop. I've been running jack
0.94.0 at 64x2 since I rebooted back into 2.4.25 with no xruns (well,
one "delay of ... exceeds spare time"). Under 2.6.4 I was getting
regular xruns at 64x2 and sporadic ones at 128x2 (hmmm, now that I think
about it, I did not verify that the process was running SCHED_FIFO, I
assume it was doing that because I specified "-R" and jackstart did not
complain). 

So, if you are concerned with good low latency performance then it is
not yet time to switch to 2.6.x[*]. 

It should be noted that I would still have to create custom kernels for
2.6.x (something I was hoping to avoid). Preempt does not seem to be
enabled by default in the configuration I tested and there is an option
that has to be changed to compile as a module to get the LSM addition to
build. 

I should try to get some hard numbers with the latencytest program... 
-- Fernando

[*] too bad for the Athlon64 platform... no lowlat patches there so
2.6.x would seem to be the only choice :-(