Re: [FFADO-devel] [MOTU] performances and low latency

[Jonathan W. <jw...@ph...>]

Hi

> > > As I can remember, it was a few versions around the last streaming
> > > rebuild (about r780) with mush better streaming stability.
> 
> r780 is not the right number. I just had a look and the streaming rework I'm
> talking about is around r734 (merge ppalmers-streaming branch).

Yep, that's the streaming rework from around the time of the freebob-ffado
rename.  We're talking about the same thing.

> > I think it's relative.
> Not so. I'm talking about big differences with the same ardour session.

One of the problems I've noticed with freebob/ffado is that performance does
seem to be tied to tiny details of machine configuration.  Certainly the
faster the machine the better it runs but there are also other issues at
play I think (irq routing, sharing and the like).  In the past I've
experienced few issues running configurations which have caused other
people's systems to give up.  This is what I meant by "relative" although I
admit I was far from clear.

> > > Those versions had other bugs especially on startup and shutdown
> > 
> > Yes.  I think Pieter's latest changes have finally sorted most of those
> > out.
> I confirm.

That's good to know.  We needed to add a few more steps to the streaming
bring-up method to allow for some of the more unusual requirements of the
MOTU devices.

> > I'll defer final comment to Pieter since he is most familiar with the
> > low level streaming infrastructure and its status.  My understanding is
> > that there are still some known low-latency issues in the low level code
> > which he is working on as time allows.
>
> According svn logs, last streaming improvements have not been propagated to
> the MOTU stream processor.

Which changeset are you referring to here?  Pieter indicated to me that he
was propagating relevant changes to MOTU so I've been paying only slight
attention to that side of things.  It's possible that time hasn't allowed a
complete update yet though.  

On the other hand, note that some of Pieter's work isn't applicable to the
MOTU, and a lot of the more recent work has been in the common
infrastructure too which hasn't required explicit changes in the driver
specific code.

> > Out of interest, how low can you go on rev 900/901/902?
> 
> With my ardour test session, it is difficult to go under 3 period/buffer and
> 512 frames/period.
> 2 periods/buffer doesn't work at all with it.

For interest and comparison I'll see if I can check this out myself tonight.

As a final comment I'll just add that over the past 2 weeks Pieter seems to
have uncovered some rather odd behaviour in the kernel ohci1394 driver.
Under certain conditions it looses packets (or at least never passes them to
userspace) and if true this may be contributing to some of the ffado
instabilities we've been seeing at low latencies.  As I understand it,
testing and categorisation is still underway.  You can pick up on this issue
in the ieee1394 mailing list.

Regards
  jonathan