On 02/14/2012 08:55 PM, Brian Monroe wrote:
On Tue, Feb 14, 2012 at 12:21 PM, Fernando Lopez-Lezcano
<nando(a)ccrma.stanford.edu <mailto:firstname.lastname@example.org>> wrote:
On 02/14/2012 11:31 AM, Brian Monroe wrote:
I've been spending a lot of time on the #opensourcemusicians channel
talking to Ubuntu Studio users about their kernel and latency times
they're getting. Seems like most of them are using g a stock
the preemptive option enabled and they are getting great latency
(2ms)while utilizing the @audio group on their user. I ended up
compiling my own low latency kernel and I haven't had any issues
yet. If this is what we are missing for the spin I'd be happy to
maintain packaging for the kernel. I know ccrma has been behind
a few kernel releases.
The latest I have, current in testing is 3.2.2 + rt11 (for Fedora 15
and 16). I am currently trying to build 3.2.6 + rt12.
Ah, that may have been an error on my part, I thought last I checked,
which admittedly was a week and a half ago there was only a F15 kernel
on the CCRMA website.
The web site needs updating (and I just released 3.2.6 rt12 for fc16
testing)... Pretty much everything that is available for fc15 is
available for fc16 (I have been using the mailing lists to announce
stuff and never have time to update the web site). Install
planetccrma-repo and that should get you going.
The current rt not in testing is a 3.0.x based release (fc15/16).
have not seen a big interest on being up to date - I just try to
keep up with the latest rt patch set. If there is more interest I
could try to keep up (but keeping up with _what_?, for a bit I was
testing a 3.2 based rt patched kernel and that was still not
available for fc16 as an official release).
I saw the instructions for adding the real time patch for a tick
kernel and from what I can tell it wouldn't be hard to get that
I'm not entirely sure what ccrma does differently with their kernels
compared to other Linux users,
"compared to other Linux users"? I don't follow.
Namely the Ubuntu Studio folks. Most users in #opensourcemusicians seem
to use Ubuntu. Why? I don't know.
Probably part of the reason is that not being a US based distribution it
does not care about some stuff that makes Fedora harder to use (the
codec patent minefield). Fedora also has a reputation of being bleeding
edge, more so than Ubuntu. As a technology testing ground it has in the
past made decisions that advanced the state of the art (although some
users question whether what happened was an "advance") and broken
important audio stuff as a side effect, sometimes for a long time. That
chases users away. I can atest to that.
and I'm still a bit of a noob so I could
be off base with this, but I would reason that we should be able
utilize the same settings to archive similar performance
I thought I read that ccrma uses a unique scheduler, but if we
a 2ms latency time without it, the point may be moot.
Nope, no unique scheduler or other stuff. Where did you read that?
To be honest I'm not sure where I read this, but I do remember having
conversations about it in one of the channels. Part of the reason I
wanted to email the list was to hear what's what from the source, so
thanks for clearing that up for me.
The Planet CCRMA rt patched kernels are based on recent Fedora
source packages (usually from Koji) that are the closest I can find
to the kernel releases for which rt patches are available. To that
source package I add the rt patch, drop Fedora patches that are
already included or conflict, and built that. I use pretty much the
stock Fedora kernel configuration files except for whatever tweaks
are necessary to enable the rt patch for full preemption. That's
Is there any help needed for testing/ect? I'm just trying to figure out
what I can do to start contributing.
If you are a Fedora user and you are interested in low latency kernels
you can of course try the Planet CCRMA kernel-rt and kernel-rtPAE
packages I maintain (plus rtirq and other goodies, install
"planetccrma-core" or "planetccrma-core-PAE"). BTW, it is only
that 3.0 and 3.2 kernels have evolved to a state where they behave as
well or better than good'old 2.6.33.x, at leaste IMO.
As for an audio oriented spin that is _really_ good, the only way that
can happen officially in Fedora (AFAIK) is for the upstream kernel to
become good enough for that without the rt patches. It is probably
getting there in terms of source code, but for the Fedora kernel to
become better than average it would need to be rebuilt with more
appropriate configuration options, namely threaded irq's.