22 Jan
2003
22 Jan
'03
10:52 a.m.
On Wed, Jan 22, 2003 at 03:22:42 +0100, Robert Jonsson wrote:
Atleast I have noticed that performance is worse
running with jack than
running without it. My evidence is basically running MusE with alsa vs
jack output. Running with jack it is much more prone to produce xruns.
I have not talked so much about this because:
a) jack is in development
Well, technically, but I regard it as stable in its behaviour. Paul, Kai?
b) I thought it was common knowledge
It seems not :)
c) I thought that because the extra weight(scheduling
etc) that jack
adds this was perfectly justified.
My system isn't perfectly tuned and that might be the reason I notice
the difference...
It shouldn't be a problem, iwht only a few apps running the overhead is
very small. If by "isn't perfectly tuned" you mean not running SCHED_FIFO
or a patched kernel that its not that supprising, the context switch will
be causing big problems. But, at worst it should just make xruns a bit more
frequent, not go from none to some.
IMHO any runtime xruns at all makes audio software unusable.
- Steve
22 Jan
22 Jan
11:04 a.m.
Hi,
What I have noticed is subtle, I have xruns in both situations but the
rate increases noticeably.
I've been running patched kernels but there are all kinds of parameters
(as I'm sure you know) to tweak before it is sufficiently solid.
I would consider my findings pretty much expected behaviour with a less
then perfectly tuned system.
The real problem is that it's so very very hard to tune a system
sufficiently without hours and hours of work. It would be very nice if
there was some way to automate, atleast parts of, the process.
/Robert
c) I thought
that because the extra weight(scheduling etc) that jack
adds this was perfectly justified.
My system isn't perfectly tuned and that might be the reason I notice
the difference...
It shouldn't be a problem, iwht only a few apps running the overhead is
very small. If by "isn't perfectly tuned" you mean not running SCHED_FIFO
or a patched kernel that its not that supprising, the context switch will
be causing big problems. But, at worst it should just make xruns a bit more
frequent, not go from none to some.
11:19 a.m.
On Wed, Jan 22, 2003 at 03:57:40 +0100, Robert Jonsson wrote:
What I have noticed is subtle, I have xruns in both
situations but the
rate increases noticeably.
I've been running patched kernels but there are all kinds of parameters
(as I'm sure you know) to tweak before it is sufficiently solid.
I would consider my findings pretty much expected behaviour with a less
then perfectly tuned system.
The real problem is that it's so very very hard to tune a system
sufficiently without hours and hours of work. It would be very nice if
there was some way to automate, atleast parts of, the process.
I didn't have to put any real effort in on my PC (RH8, ide, athlon),
I just installed a patched kernel from Planet CCRMA and rebooted.
I get xruns at 64 samples/period (and 128 with heavy disk load), but I
think that's a problem with the ext3 filesystem - others have reported this
too. I'm happy at 256 so I haven't bothered remounting the partition to
find out.
JACK/no JACK makes no noticable difference to the xruns.
- Steve
11:34 a.m.
It
shouldn't be a problem, iwht only a few apps running the overhead is
very small. If by "isn't perfectly tuned" you mean not running SCHED_FIFO
or a patched kernel that its not that supprising, the context switch will
be causing big problems. But, at worst it should just make xruns a bit more
frequent, not go from none to some.
What I have noticed is subtle, I have xruns in both situations but
the rate increases noticeably. I've been running patched kernels but
there are all kinds of parameters
(as I'm sure you know) to tweak before it is sufficiently solid. I
would consider my findings pretty much expected behaviour with a
less then perfectly tuned system. 
12:22 p.m.
On Wed, Jan 22, 2003 at 04:26:02PM +0100, Dave Griffiths wrote:
I thought I'd pipe up here, as I use jack on a
totally unpatched, untuned
system (with SCHED_FIFO) and get rock solid performance. All crashes etc have
so far been problems with my code :)
It should provide to be useful if we could compile a list
of hardware & software (kernel, patches, distro, sysctl ...) settings
which result in rock solid performance for jack (maybe we could
specify the buffer size too).
I think someone was already thinking of doing sth. like this ?
v
8050
days inactive
8050
days old
linux-audio-dev@lists.linuxaudio.org
4 comments
4 participants
participants (4)
-
Dave Griffiths -
Robert Jonsson -
Steve Harris -
Vincent Touquet