24 Mar
2010
24 Mar
'10
8:51 p.m.
(forgot to copy this to LAD)
On Wed, Mar 24, 2010 at 06:45:00PM -0400, Paul Davis wrote:
On Wed, Mar 24, 2010 at 5:59 PM,
<fons(a)kokkinizita.net> wrote:
It was -X raw. I just repeated the test with -X seq
and that provides a completely different picture.
Jitter is +/- 2 frames, with occasional outliers
at around 160 frames, nothing in between.
That is with a 30-track Ardour session running in
the background, DSP load 15%, CPU load 27%.
Unpatched 2.6.32 kernel (which may explain the
occasional larger error).
Not bad at all.
Ciao,
--
FA
O tu, che porte, correndo si ?
E guerra e morte !
When connected via a loopback on a HW interface
I expected the worst case to be events quantised
to Jack period (256 frames). Actually it's 10 times
what bridge were you usng between JACK and ALSA? 
25 Mar
25 Mar
6:19 a.m.
Excerpts from fons's message of 2010-03-25 00:47:08 +0100:
(forgot to copy this to LAD)
On Wed, Mar 24, 2010 at 06:45:00PM -0400, Paul Davis wrote:
Very strange, -Xraw is said to perform better than -Xseq.
Maybe there's a difference between outboard and software?
In software land jack midi is the clear winner, -Xraw is said to perform
better than -Xseq, so if it's different for outboard gear, where does
the difference come from?
On Wed, Mar 24, 2010 at 5:59 PM,
<fons(a)kokkinizita.net> wrote:
It was -X raw. I just repeated the test with -X seq
and that provides a completely different picture.
Jitter is +/- 2 frames, with occasional outliers
at around 160 frames, nothing in between.
That is with a 30-track Ardour session running in
the background, DSP load 15%, CPU load 27%.
Unpatched 2.6.32 kernel (which may explain the
occasional larger error).
Not bad at all.
Ciao, When connected via a loopback on a HW interface
I expected the worst case to be events quantised
to Jack period (256 frames). Actually it's 10 times
what bridge were you usng between JACK and ALSA?
7:27 a.m.
On Thu, Mar 25, 2010 at 10:19:50AM +0100, Philipp wrote:
Very strange, -Xraw is said to perform better than
-Xseq.
Maybe there's a difference between outboard and software?
In software land jack midi is the clear winner, -Xraw is said to perform
better than -Xseq, so if it's different for outboard gear, where does
the difference come from?
I haven't looked closely at the MIDI code in Jack's Alsa
backend, so I can't tell. But AFAIK
1. Alsa's Raw interface does not provide scheduling of
events to be transmitted at some particular time,
nor timestamps on received events.
If timing is important to your use, you have to do
it yourself, by calling the write() function at the
right time, and by reading some timer for received
data.
Assuming that the backend is not doing this, event
timing would be quantised to audio period boundaries.
Note that doing your own timing in this case requires
a separate thread - the backend as seem from Jack's
engine must never wait, except for audio period
boundaries.
Also to ensure immediate transmission when calling
the write() function, the caller has to either open
the device with the SYNC flag set, or use the drain()
function. Failure to do either could explain the very
large quantisation I observed with the raw interface
- it leaves it up to the Alsa code when to transfer
data from user space to kernel (driver) space, and
this decision will probably be based on the number
of bytes waiting etc.
2. The Alsa Sequencer interface allows to put a timestamp
on transmitted data, and then Alsa will ensure it gets
transmitted at the right time. It will also provide
timestamps on received data.
Ciao,
--
FA
O tu, che porte, correndo si ?
E guerra e morte !
26 Mar
26 Mar
5:47 a.m.
Excerpts from fons's message of 2010-03-25 11:22:40 +0100:
On Thu, Mar 25, 2010 at 10:19:50AM +0100, Philipp
wrote:
Thanks Fons.
It's a bit of technical detail overload for me, but in essence, if I got
it right:
Alsa sequencer does timestamping/scheduling.
Jack midi does it as well, probably more accurate.
Raw doesn't do it at all.
Consequences:
..there are so many possible combinations here, especially with
a2jmidid in the equation, and they'd need to be tested :)
Very strange, -Xraw is said to perform better
than -Xseq.
Maybe there's a difference between outboard and software?
In software land jack midi is the clear winner, -Xraw is said to perform
better than -Xseq, so if it's different for outboard gear, where does
the difference come from?
I haven't looked closely at the MIDI code in Jack's Alsa
backend, so I can't tell. But AFAIK
1. Alsa's Raw interface does not provide scheduling of
events to be transmitted at some particular time,
nor timestamps on received events.
If timing is important to your use, you have to do
it yourself, by calling the write() function at the
right time, and by reading some timer for received
data.
Assuming that the backend is not doing this, event
timing would be quantised to audio period boundaries.
Note that doing your own timing in this case requires
a separate thread - the backend as seem from Jack's
engine must never wait, except for audio period
boundaries.
Also to ensure immediate transmission when calling
the write() function, the caller has to either open
the device with the SYNC flag set, or use the drain()
function. Failure to do either could explain the very
large quantisation I observed with the raw interface
- it leaves it up to the Alsa code when to transfer
data from user space to kernel (driver) space, and
this decision will probably be based on the number
of bytes waiting etc.
2. The Alsa Sequencer interface allows to put a timestamp
on transmitted data, and then Alsa will ensure it gets
transmitted at the right time. It will also provide
timestamps on received data.
Ciao,
--
FA
O tu, che porte, correndo si ?
E guerra e morte ! 
7:23 a.m.
On Fri, Mar 26, 2010 at 4:46 AM, Philipp <hollunder(a)lavabit.com> wrote:
al detail overload for me, but in essence, if I got
it right:
Alsa sequencer does timestamping/scheduling.
not quite. it *can* do timestamping/scheduling. it doesn't have to,
and a2jmidid doesn't use it that way.
Jack midi does it as well, probably more accurate.
JACK MIDI is sample accurate.
Raw doesn't do it at all.
but in theory has a slightly faster code path from the app to the wire.
--p
9:19 a.m.
On Fri, Mar 26, 2010 at 06:23:49AM -0400, Paul Davis wrote:
On Fri, Mar 26, 2010 at 4:46 AM, Philipp
<hollunder(a)lavabit.com> wrote:
not quite. it *can* do timestamping/scheduling. it doesn't have to,
and a2jmidid doesn't use it that way.
What is the advantage of using the seq interface
if you don't use the scheduling ? (assuming you
only want to talk to the HW ports)
Ciao,
--
FA
O tu, che porte, correndo si ?
E guerra e morte !
9:24 a.m.
On Fri, Mar 26, 2010 at 8:14 AM, <fons(a)kokkinizita.net> wrote:
On Fri, Mar 26, 2010 at 06:23:49AM -0400, Paul Davis
wrote:
routing to other ALSA-seq based software.
that's about it.
On Fri, Mar 26, 2010 at 4:46 AM, Philipp
<hollunder(a)lavabit.com> wrote:
not quite. it *can* do timestamping/scheduling. it doesn't have to,
and a2jmidid doesn't use it that way.
What is the advantage of using the seq interface
if you don't use the scheduling ? (assuming you
only want to talk to the HW ports) 
12:49 p.m.
Paul Davis wrote:
On Fri, Mar 26, 2010 at 8:14 AM,
<fons(a)kokkinizita.net> wrote:
> On Fri, Mar 26, 2010 at 06:23:49AM -0400, Paul Davis wrote:
>
>
>> On Fri, Mar 26, 2010 at 4:46 AM, Philipp <hollunder(a)lavabit.com> wrote:
>>
>> not quite. it *can* do timestamping/scheduling. it doesn't have to,
>> and a2jmidid doesn't use it that way.
>>
> What is the advantage of using the seq interface
> if you don't use the scheduling ? (assuming you
> only want to talk to the HW ports)
Paul,
could you add Fons jitter tests to JACK1 and who ever is responsible for
JACK2, could you add it to JACK2?
It would be good to have those tests as default tests on music live
media. So it would be very easy to test e.g. USB MIDI interfaces on
Windows computers of friends too. I guess this would be the only way to
have the apps guaranteed on music live media in the future. Imagine a
situation, somebody goes to buy an interface in a music store. Perhaps
they would allow somebody they know to use a live media on a computer of
the store. For now I'll solve this by storing his apps on an USB stick.
Ralf
5306
days inactive
5308
days old
linux-audio-dev@lists.linuxaudio.org
7 comments
4 participants
participants (4)
-
fons＠kokkinizita.net -
Paul Davis -
Philipp -
Ralf Mardorf