10 Nov
2005
10 Nov
'05
9:05 p.m.
Hello,
I have given up on the mainstream Fedora 4 kernel and am trying to build
2.6.14-rt5. So far I can get it built and running, but without jack it
seems.
I am wondering what are the best settings for jack using Ingo's RT
patch: how do I make the most of the patch?
(I am trying to use CONFIG_PREEMPT_RT=y)
I use qjackctl and my .jackrc is: /usr/local/bin/jackd -u -dalsa -dhw:1
-r48000 -p128 -n2 -m -H -M
hw:1 corresponds to the card number from /proc/asound/cards associated
with my RME Hammerfall Multiface.
I may have to change my .asoundrc as in the mainstream kernel cards 0
and 1 are switched for some reason (if that is what is sotpping jackd).
The only message I get is something to the effect of NTpolltime timed
out, or something... My biggest question is not how to get jackd working
- I will fix that. I am curious how to get PD running on my machine
processing complex patches but with a latency of 5.33 msec (128 frame
buffer size I think) or less. I would even settle fo 10msec (256) with
no XRUNS.
Thank you all for your help!
-thewade
10 Nov
10 Nov
9:12 p.m.
New subject: [linux-audio-user] Ingo's RT patch: please tell this newbie the best jack settings
On Thu, 2005-11-10 at 17:05 -0800, thewade wrote:
I am curious how to get PD running on my machine
processing complex patches but with a latency of 5.33 msec (128 frame
buffer size I think) or less. I would even settle fo 10msec (256) with
no XRUNS.
You don't need the -rt kernel for that, the standard kernel will work.
Lee
11 Nov
11 Nov
7:22 a.m.
New subject: [linux-audio-user] Ingo's RT patch: please tell this newbie the best jack settings
On Thu, 10 Nov 2005 20:10:20 -0500
Lee Revell <rlrevell(a)joe-job.com> wrote:
On Thu, 2005-11-10 at 17:05 -0800, thewade wrote:
Oh well, not for everyone. If you want to be sure and have no xrun
whatsoever a -rt kernel is definetly the better choice imho (don't have
data to back that up though), especially because of the irq handler
prios, which are not available on vanila kernels (maybe that has changed
though). Seeting up the priorities right is crucial in setting up an
audio system. I.e. i can run buffersizes of 32 or even 16 frames w/o
xruns (except for application provoked ones), but of course the context
switch overhead wastes quite a few cycles at these insane settings. But
it's a good test case so one can be sure everything will be smooth at 64
frames.
Have fun,
Flo
--
Palimm Palimm!
http://tapas.affenbande.org
I am curious how to get PD running on my machine
processing complex patches but with a latency of 5.33 msec (128 frame
buffer size I think) or less. I would even settle fo 10msec (256) with
no XRUNS.
You don't need the -rt kernel for that, the standard kernel will work.
Lee
3:19 p.m.
New subject: [linux-audio-user] Ingo's RT patch: please tell this newbie the best jack settings
On Fri, 2005-11-11 at 12:22 +0100, Florian Schmidt wrote:
On Thu, 10 Nov 2005 20:10:20 -0500
Lee Revell <rlrevell(a)joe-job.com> wrote:
I do have some data to back this up, I ran about 100 million samples
with 2.6.14 and the biggest latency I saw was ~0.6 ms. For the buffer
sizes he wants to use 2.6.14 will work fine. Of course you won't be
able to sustain quite as high a DSP load but given the problems people
are having with the -rt kernel on x64 it's worth investigating.
Lee
On Thu, 2005-11-10 at 17:05 -0800, thewade
wrote:
Oh well, not for everyone. If you want to be sure and have no xrun
whatsoever a -rt kernel is definetly the better choice imho (don't have
data to back that up though) I am curious how to get PD running on my machine
processing complex patches but with a latency of 5.33 msec (128 frame
buffer size I think) or less. I would even settle fo 10msec (256) with
no XRUNS.
You don't need the -rt kernel for that, the standard kernel will work.
Lee
12:04 a.m.
On 11/10/05, thewade <pdman(a)aproximation.org> wrote:
Hello,
I have given up on the mainstream Fedora 4 kernel and am trying to build
2.6.14-rt5. So far I can get it built and running, but without jack it
seems.
I am wondering what are the best settings for jack using Ingo's RT
patch: how do I make the most of the patch?
(I am trying to use CONFIG_PREEMPT_RT=y)
I use qjackctl and my .jackrc is: /usr/local/bin/jackd -u -dalsa -dhw:1
-r48000 -p128 -n2 -m -H -M
hw:1 corresponds to the card number from /proc/asound/cards associated
with my RME Hammerfall Multiface.
I may have to change my .asoundrc as in the mainstream kernel cards 0
and 1 are switched for some reason (if that is what is sotpping jackd).
The only message I get is something to the effect of NTpolltime timed
out, or something... My biggest question is not how to get jackd working
- I will fix that. I am curious how to get PD running on my machine
processing complex patches but with a latency of 5.33 msec (128 frame
buffer size I think) or less. I would even settle fo 10msec (256) with
no XRUNS.
Thank you all for your help!
-thewade
Hi,
A few comments:
1) The current kernel from Ingo would be 2.4.16-rt9. I think that rt5
didn;t last very long and had some problems. Please try -rt9.
2) Make sure you reinstall the realtime-lsm package after building that kernel.
3) Make sure you create a realtime group and modprobe realtime to
enable that group:
modprobe realtime gid=600 any=1
where gid=600 is for my realtime group in /etc/groups.
4) Making best use of the kernel requires that you understand the ISR
number for your sound card and reset it's priority higher. This helps
a lot in reducing or eliminating xruns.
5) Remove .jackrc completely and let your system set a new one.
6) Here are my HDSP 9652 settings:
jackd -R -p 64 -d alsa -r 44100 -p 64 -n2 -d hw:1,0
I get a few xruns at -p 64 -n2. I get none at -p 256 -n 2. I am still
looking for the root cause of the few xruns I get at 64/2.
7) Make sure you have the Jack FIFOS mounted on tmpfs.
As a comment, I don't personally like Jack in /usr local. You might
consider rebuilding Jack one of these days with:
./configure --prefix=/usr
Hope this all helps a bit,.
Cheers,
Mark
7:13 a.m.
On Thu, 10 Nov 2005 20:03:59 -0800
Mark Knecht <markknecht(a)gmail.com> wrote:
jackd -R -p 64 -d alsa -r 44100 -p 64 -n2 -d hw:1,0
jackd -R -P 70 ...
to set the priority of jackd higher than all soundcard irq handlers
besides the soundcard irq.
Basically the priority setup should look something like this:
keyboard irq handler - 99
rtc irq handler - 99
soundcard irq handler - 98
jackd prio - 70 (creates watchdog at prio 80 and client
process() threads at prio 69)
all other irq handlers - around 50 (default)
Basically there's also some problems with the timer irq's on -rt kernels
(which my rt watchdog suffers from). Usually you would want to set
softirq-timer and softirq-ktime to high prios, too (so high prio
sleeping threads preempt low prio running threads when getting woken up.
Actually this should make sure that the timer interrupt which wakes up
the high prio thread after its sleep actually gets to run), but this
doesn't quite work with the kernels i tested (this is the reason why my
rt_watchdog doesn't work right).
I'll try with newer kernels and ask ingo about it if the problem
persists. Anyways, this should really only be a problem with midi
sequencers that are sleep based. When using an irq source for timing
this problem should go away.
Regards,
Flo
--
Palimm Palimm!
http://tapas.affenbande.org
7:29 a.m.
On Friday 11 November 2005 09:13 am, Florian Schmidt wrote:
would this help my jackd run more smoothly? (now i have lots of interruptions
while playing midis)
On Thu, 10 Nov 2005 20:03:59 -0800
Mark Knecht <markknecht(a)gmail.com> wrote:
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
jackd -R -p 64 -d alsa -r 44100 -p 64 -n2 -d
hw:1,0
jackd -R -P 70 ...
to set the priority of jackd higher than all soundcard irq handlers
besides the soundcard irq.
Basically the priority setup should look something like this:
keyboard irq handler - 99
rtc irq handler - 99
soundcard irq handler - 98
jackd prio - 70 (creates watchdog at prio 80 and client
process() threads at prio 69)
all other irq handlers - around 50 (default)
Basically there's also some problems with the timer irq's on -rt kernels
(which my rt watchdog suffers from). Usually you would want to set
softirq-timer and softirq-ktime to high prios, too (so high prio
sleeping threads preempt low prio running threads when getting woken up.
Actually this should make sure that the timer interrupt which wakes up
the high prio thread after its sleep actually gets to run), but this
doesn't quite work with the kernels i tested (this is the reason why my
rt_watchdog doesn't work right).
I'll try with newer kernels and ask ingo about it if the problem
persists. Anyways, this should really only be a problem with midi
sequencers that are sleep based. When using an irq source for timing
this problem should go away.
Regards,
Flo
3:45 p.m.
New subject: [linux-audio-user] Ingo's RT patch: please tell this newbie the best jack settings
On Fri, 2005-11-11 at 12:13 +0100, Florian Schmidt wrote:
On Thu, 10 Nov 2005 20:03:59 -0800
Mark Knecht <markknecht(a)gmail.com> wrote:
Before you go nuts with tweaking all this stuff, see if 2.6.14 works for
you. It might Just Work. I'm having great results with it here.
Lee
jackd -R -p 64 -d alsa -r 44100 -p 64 -n2 -d
hw:1,0
jackd -R -P 70 ...
to set the priority of jackd higher than all soundcard irq handlers
besides the soundcard irq.
9:55 p.m.
On 11/11/05, Lee Revell <rlrevell(a)joe-job.com> wrote:
On Fri, 2005-11-11 at 12:13 +0100, Florian Schmidt
wrote:
I can empirically support Lee's statements. I run a plain
gentoo-sources 2.6.13 kernel on all of my 32-bit machines. I have no
problems even with Jack running at 64/2.
It's only on my AMD64 machine that I've had to use the -rt patches. I
suspect that when the last of the problems get worked out of the
64-bit kernels (or NIC drivers, or video drivers, etc.) I may not need
to use the -rt patches for that machine either.
As Lee says - try it. It just may work.
- Mark
On Thu, 10 Nov 2005 20:03:59 -0800
Mark Knecht <markknecht(a)gmail.com> wrote:
Before you go nuts with tweaking all this stuff, see if 2.6.14 works for
you. It might Just Work. I'm having great results with it here.
Lee jackd -R -p 64 -d alsa -r 44100 -p 64 -n2 -d
hw:1,0
jackd -R -P 70 ...
to set the priority of jackd higher than all soundcard irq handlers
besides the soundcard irq.
6954
days inactive
6955
days old
linux-audio-user@lists.linuxaudio.org
8 comments
5 participants
participants (5)
-
Florian Schmidt -
I. I. Ooisen -
Lee Revell -
Mark Knecht -
thewade