[PlanetCCRMA] Latencies in 2.6 vs 2.4 (was: PlanetCCRMA)

[Fernando Lopez-Lezcano]

On Wed, 2004-11-24 at 19:23, Axel Thimm wrote:
> On Wed, Nov 24, 2004 at 09:05:50PM -0500, Gene Heskett wrote:
> > On Wednesday 24 November 2004 20:31, Axel Thimm wrote:
> > >On Wed, Nov 24, 2004 at 12:46:25PM +0100, R.M.Deal wrote:
> > >> Does anyone have any idea when PlanetCCRMA, the source for audio
> > >> software from
> > >> Stanford University, will run smoothly in FC3?  With RH9, it was
> > >> an excellent way to
> > >> keep up with developments tracked/developed at Stanford Univ.,
> > >> using apt-get.

I haven't yet started the rebuild of all packages for fc3, sorry. Even
when I get to do that (soon), the kernel may still be the main stumbling
block for a good pro audio fc3 platform (_and_ fc2, for that matter). 

> > >> If this is a FAQ, please point me to the appropriate archive.  I
> > >> just installed FC3.

You could use the fc2 repository in the meanwhile, I guess. I know for
sure that apps that depend on libcurl will not work (unless a backward
compatibility package is created - the library version has changed and
is not compatible with the old one). 

> > >Currently PlanetCCRMA is still recommending using FC1 (with
> > >PlanetCCRMA's 2.4 kernel) over any 2.6 kernel based distribution for
> > >the still unmatched latency features.

_And_ stability. 

> > >>   Ralph M. Deal, retired prof. of Phys. Chem.
> > 
> > That recommendation may not be so ironclad here shortly.
> 
> Everybody hopes so :)

Me too...

> > I'm running Ingo Molnar's morning patch at full preemption setting
> > here, uptime is about 7 hours now, no glitches as I go about my
> > normal daily routines, and its obviouly much snappier than a home
> > built 2.6.10-rc2-mm3 kernel.
> 
> Sounds nice, but prof. audio apps have rather rigid latency
> demands. It would be helpful to measure latencies and report on the
> PlanetCCRMA lists (if you are audio/sound interested at PlanetCCRMA's
> level). Fernando Pablo Lopez, the maintainer of PlanetCCRMA, could
> outline or even guide you through the benchmarking.
> 
> Moving this to the PlanetCCRMA lists, as this is more appropriate
> there.
> 
> > I keep waiting for the other shoe to drop, and it may well, but so
> > far it been very solid & apparently at least bullet-resistant.
> 
> > >From another mostly retired old coot.

I have been trying (rather unsuccessfully) to keep up with the frantic
release of new patches by Ingo and all the other helpful gurus and
non-gurus that contribute, but so far I have not had that much luck. At
least not something I can recommend without reservations for pro audio
work. 

I do have a couple of "test" kernels in the Planet Edge repository that
have had some success in fc2 (based on R9 and S7). The "voluntary
preemption" series was approaching stability, and some users were very
happy with the results they were getting, but then the "realtime
preemption" appeared, with potentially even better performance. I'm sure
this will become a reality soon, but at this point I can't even get a
completely clean boot on my test machines with the latest patchsets
(0.7.31-3 this morning, I think), and the realtime performance, at least
in the hardware I have tested on, is not stellar. Again, this is to be
expected. 

A caveat here, what I mainly use to test is Jack (the Jack Audio
Connection Kit, a very low latency sound server) and its most popular
applications, usually running with 2 buffers of 128 samples (or even
64). Jack uses realtime priority for its audio threads (and the audio
threads of the client applications) to ensure dropout free performance
at low latencies (and for that needs an additional kernel module to be
able to be run from regular non-root accounts with SCHED_FIFO). There
may be even bugs in Jack that are being highlighted by the better
potential performance of 2.6.x...

It all depends on your requirements. For regular audio work the current
situation is probably good enough. For very low latency without dropouts
in a pro environment it is not yet there in my experience (again,
depends on the exact hardware you have as well). Keep in mind that you
don't want _any_ dropouts for the kind of work I'm targeting, even with
a fully loaded machine (and no, 2.4.x plus all the low latency goodies
is not absolutely dropout free either at very very low latencies - the
potential is there in 2.6.x to be much better!). 

Just a comment...
I _want_ to be proved wrong, _very_ soon!
-- Fernando