8 Oct
2009
8 Oct
'09
5:25 a.m.
Hello,
I suppose this question has already been posted, and answered, but I
can't get any explanation which seems to fit my question, or I do not
understand it, so I post it there.
I know that Jack DSP Load isn't a true DSP Load, it's nothing but
buffering, if I'm not wrong.
Though, I don't know what hardware part of the computer the DSP finally
relies to (or affects, or whatever), but I suppose it's the processor,
as the load of the DSP when launching, for example, one AMS seems to be
the same as the processor load assigned to the AMS process.
So, the first question is, does the DSP mean "processor load due to
software appearing in jack"?
Next, question is about multicore. When I monitor my system using top on
a my quad-core or duo-core machines, I can reach total processor loads
such as 400% or 200%, as top seems to be designed for monocore.
Actually, it obviously means 4x100% or 2x100%.
When I use Jack1, I got an Xrun (at least) every time I reach the 100%
DSP load, which is finally with quite few softwares (4 AMS, 1
Sooperlooper, 1 seq24, 1 Tapeutape, some routing, some qmidiroute) for
my configuration (Intel Q6600 Core 2 Quad, 8Gb RAM, Focusrite Saffire
Pro 40 soundcard -firewire, with developpment version of FFADO-).
What I was wondering is: does jack prefer to create an Xrun as soon as
100% DSP load is reached?
I just installed jack2, in order to check whether it might increase the
perf of my system, but, at first sight, the DSP load is the very same as
with jack1.
OK, here it is, I'm sorry it's not really clearly explained, it's quite
clear in my head, but unfortunately, with french words!!!
Thanks.
--
Aurélien
11 Oct
11 Oct
10:17 a.m.
Perhaps my question is not so-well posed.
Let's turn it like this:
How is the DSP load appearing in jack computed, and how does it have to
deal with CPU/Memory/whateverphysicalthing load?
Does jack generate an xrun as soon as it reaches 100% DSP load as a
"security operation" or is it due to hardware limitation when reaching
100% DSP load?
Thanks.
On Thu, Oct 08, 2009 at 09:25:18AM +0200, Aurelien wrote :
Hello,
I suppose this question has already been posted, and answered, but I
can't get any explanation which seems to fit my question, or I do not
understand it, so I post it there.
I know that Jack DSP Load isn't a true DSP Load, it's nothing but
buffering, if I'm not wrong.
Though, I don't know what hardware part of the computer the DSP finally
relies to (or affects, or whatever), but I suppose it's the processor,
as the load of the DSP when launching, for example, one AMS seems to be
the same as the processor load assigned to the AMS process.
So, the first question is, does the DSP mean "processor load due to
software appearing in jack"?
Next, question is about multicore. When I monitor my system using top on
a my quad-core or duo-core machines, I can reach total processor loads
such as 400% or 200%, as top seems to be designed for monocore.
Actually, it obviously means 4x100% or 2x100%.
When I use Jack1, I got an Xrun (at least) every time I reach the 100%
DSP load, which is finally with quite few softwares (4 AMS, 1
Sooperlooper, 1 seq24, 1 Tapeutape, some routing, some qmidiroute) for
my configuration (Intel Q6600 Core 2 Quad, 8Gb RAM, Focusrite Saffire
Pro 40 soundcard -firewire, with developpment version of FFADO-).
What I was wondering is: does jack prefer to create an Xrun as soon as
100% DSP load is reached?
I just installed jack2, in order to check whether it might increase the
perf of my system, but, at first sight, the DSP load is the very same as
with jack1.
OK, here it is, I'm sorry it's not really clearly explained, it's quite
clear in my head, but unfortunately, with french words!!!
Thanks.
--
Aurélien
_______________________________________________
Linux-audio-user mailing list
Linux-audio-user(a)lists.linuxaudio.org
http://lists.linuxaudio.org/mailman/listinfo/linux-audio-user
--
Aurélien
12:15 p.m.
On Sun, Oct 11, 2009 at 02:17:30PM +0200, Aurelien wrote:
Perhaps my question is not so-well posed.
Let's turn it like this:
How is the DSP load appearing in jack computed, and how does it have to
deal with CPU/Memory/whateverphysicalthing load?
It is the total time taken by all clients to finish
the processing for one period divided by the period
time. The actual value returned by Jack and shown
in qjackctl and ardour is the maximum of this ratio
over the recent past.
This means that if a jack client sleeps for half a
period time in its process callback (it shouldn't
do that of course) that will show up as 50% in the
'DSP load', even if it doesn't take any real CPU.
Does jack generate an xrun as soon as it reaches 100%
DSP load as a
"security operation" or is it due to hardware limitation when reaching
100% DSP load?
The xruns you see when Jack is running on AlSA are
reported by ALSA. They are usually not related to
DSP load but could have all sorts of causes. If you
have such xruns they will typically show up even
without any Jack client being active.
When a process cycle takes too long Jack will detect
this, and AFAIK remove the client that was active
when the new period should have started. This client
may or may not be the one that takes too much time.
Ciao,
--
FA
Io lo dico sempre: l'Italia è troppo stretta e lunga.
12:33 p.m.
On Sun, Oct 11, 2009 at 04:12:05PM +0200, Fons Adriaensen wrote :
Actually, I do not get any xruns, except when load just comes to 100%. I
work with FFADO (unstable) on a Focusrite Saffire Pro 40, and have a
quad-core machine with 8Gb. I mainly use synth (alsa modular x4),
sooperlooper (2 stereo loopers, 2 mono), tapeutape and some (many) audio
routing. I usually work around 80-85% (without sequencer playing) and 90
% load. But sometimes (really not often), I've got this load increase
and then an xrun. It could be widely sufficient if it wasn't purposed
to play onstage, actually.
On Sun, Oct 11, 2009 at 02:17:30PM +0200, Aurelien
wrote:
OK. (btw, how is defined the recent past?)
Perhaps my question is not so-well posed.
Let's turn it like this:
How is the DSP load appearing in jack computed, and how does it have to
deal with CPU/Memory/whateverphysicalthing load?
It is the total time taken by all clients to finish
the processing for one period divided by the period
time. The actual value returned by Jack and shown
in qjackctl and ardour is the maximum of this ratio
over the recent past.
This means that if a jack client sleeps for half a
period time in its process callback (it shouldn't
do that of course) that will show up as 50% in the
'DSP load', even if it doesn't take any real CPU.
OK.
Does jack generate an xrun as soon as it reaches
100% DSP load as a
"security operation" or is it due to hardware limitation when reaching
100% DSP load?
The xruns you see when Jack is running on AlSA are
reported by ALSA. They are usually not related to
DSP load but could have all sorts of causes. If you
have such xruns they will typically show up even
without any Jack client being active.
When a process cycle takes too long Jack will detect
this, and AFAIK remove the client that was active
when the new period should have started. This client
may or may not be the one that takes too much time.
OK.
Ciao,
--
FA
Io lo dico sempre: l'Italia è troppo stretta e lunga.
_______________________________________________
Linux-audio-user mailing list
Linux-audio-user(a)lists.linuxaudio.org
http://lists.linuxaudio.org/mailman/listinfo/linux-audio-user
--
Aurélien
1:10 p.m.
On Sun, Oct 11, 2009 at 04:32:29PM +0200, Aurelien wrote:
OK. (btw, how is defined the recent past?)
I would have to scan the sources to find out :-)
Actually, I do not get any xruns, except when load
just comes to 100%. I
work with FFADO (unstable) on a Focusrite Saffire Pro 40, and have a
quad-core machine with 8Gb. I mainly use synth (alsa modular x4),
sooperlooper (2 stereo loopers, 2 mono), tapeutape and some (many) audio
routing. I usually work around 80-85% (without sequencer playing) and 90
% load. But sometimes (really not often), I've got this load increase
and then an xrun. It could be widely sufficient if it wasn't purposed
to play onstage, actually.
If you get a reliable 85% that is actually quite good
and it would seem unwise to want more, in particular for
live work. The actual amount of work done in a synth will
not be exactly constant, it may depend on the number of
active voices, or you may have algorithms that have a
complexity that depends on the signal itself, so you
should allow for some spare time. And the rest of your
system needs to run as well !
FFADO if I'm not mistaken is multi-threaded, and I suspect
its threads could be at lower priority than Jack's. You
can easily verify this with e.g. top (use 'H' to show
all threads). So if the jack clients don't leave enough
time to FFADO, it will get an xrun. You could try lowering
Jack's priority. That will not solve the problem, but the
result should be different - no xruns but probably Jack
reporting it doesn't get enough time to run all clients,
and maybe removing one of them.
BTW, on your quad-core you should be running jack2 aka
jackdmp. With jack1 and almost all load in audio apps
you are probably not exploiting the full power of your
machine.
Ciao,
--
FA
Io lo dico sempre: l'Italia è troppo stretta e lunga.
5614
days inactive
5617
days old
linux-audio-user@lists.linuxaudio.org
4 comments
2 participants
participants (2)
-
Aurelien -
Fons Adriaensen