[Jack-Devel] Failures in connecting mac OSX Sierra and an embedded board with linux via netjack2

[Chris Caudle]

On Wed, April 26, 2017 6:24 pm, Frodo Jedi wrote:
> Actually, my goal is that of sending audio from the board to
> other connected devices, for the moment I am using a mac,
> it could be another pc with linux or windows.

Jack is designed with the idea that the audio hardware controls all data
flow, since that is really the final location that the data has to either
get to the output device (D/A converter, S/PDIF output, etc.), or is sent
by the input device (A/D converter).  In jack terms what you are
describing is that the odroid would be sent to multiple master devices. 
You either have to disconnect from one master and connect to another each
time you want to send audio to a different device, or you have to use a
bridge software to bridge between different jack master clock domains. The
jack walk-through that you linked to previously shows using the
audioadapter plugin for that, the zita-nj and zita-jn bridges work better.

>  I will try zita but my main goal is to use netjack2.

You should be very clear whether your goal is to use netjack2, or whether
your goal is to send audio from your odroid to various other devices. If
your goal is to use netjack2 you should pick a connection topology that
matches the expectations of the jack design.

If your goal is actually to send audio from the odroid to various other
devices, then you should use whatever tool is most appropriate for the
job.  If you need audio to be sent to multiple other devices with audio
hardware using jack, then zita bridge is probably what you need if you are
using jack.  Whether you should even be using jack is a separate question.

Is this a studio or audio production type setup?  Because if you are just
trying to setup distributed home audio then something like Logitech Media
Server and squeezebox lite software may be better suited for what you are
trying to do.

> I am going for a wired connection first, with the final aim to
> test netjack2 over wifi, and compare the wired and wireless
> behaviours  under various conditions.

Don't have high expectations for jack over WiFi.  Jack is designed around
low latency connections, and WiFi generally cannot sustain reliably low
latency connections.  There is work going on to get there, but unless you
have been following along with the bufferbloat and "make Wi-Fi fast" work
and are running bleeding edge drivers you are probably going to hit
latency spikes that will cause problems for jack.  You can have fun
figuring that out once wired connections are working OK.


> The odroid board (configured as master in the first case)
> has a beheringer soundcard perfectly working, I can record
> and listen audio using it. So I should be able to use the
> odroid + soundcard configuration as a master, and the
> mac as a slave. Right?

Yes, presumably.  Just to be clear, those terms mean you would have
something like Ardour running on the mac producing audio data, and the
data would be sent to the odroid which would play the audio through the
behringer/speakers.

> Ok, but my goal is to have a networked configuration where
> the linux odroid system broadcast audio to various connected devices.

With the various connected devices each having audio hardware and speakers
connected?  Then you will need bridge software such as audioadapter or
zita-njb and zita-jnb loaded to cross between the clock domains (since
each separate audio hardware will have its own sample clock running
independently and you don't have a way to synchronize all the clocks).

> My understanding is that the odroid must be the master and
> all the other devices receiving and reproducing the audio
> signal must be the slave. Am I wrong?

You are wrong.  In the jack terminology, the device with the fixed clock,
i.e. the audio hardware device, must control the flow of data, in other
words must be the master.
In the case of multiple devices playing out the audio you have in jack
terminology multiple masters, so you have to bridge between the masters.

Just for the sake of  completeness, there is a way to do what you thought
you wanted, used in studio and broadcast facilities, but it requires all
the audio playback devices to have the ability to synchronize their audio
clocks to a centralized clocking signal used for the entire studio or
broadcast facility.  Traditionally that was done by distributing a
synchronization signal to each piece of equipment used, more recently it
can be done by using IEEE 1588v2 "Precision Time Protocol" to synchronize
the local clock of each device over Ethernet.  That is a fairly
sophisticated setup, suffice to say that if you had equipment that could
do that you would already know how to set it up correctly, so that
topology does not apply here.

Jackd may not be the tool best suited for your use.  Do you actually need
low latency and the patch bay style features of jack?  If you just want to
be able to play audio from a central server to various audio playback
devices which do not need to be synchronized at the audio frame level then
some type of whole house audio design may be better suited to what you
want than jack, which is designed specifically for low latency audio
production use.  Either Logitech Media Server with squeezebox lite
instances as I mentioned, or an Icecast server running on the odroid and
VLC or something similar running on the other machines to pull the audio
they want.  That is the answer to a question you did not ask, just wanted
to mention it for completeness, since you have asked about getting netjack
running but it is not clear from some of your questions whether that is
really what would best serve your end goal (assuming your end goal really
is some arrangement of playing audio, and contrary to what you wrote
earlier running netjack2 is not actually a goal in itself).

>> You are giving a unicast address as the argument for the
>> multicast address
> I don't have the setup right now, I will re-test tomorrow,
> but if  I correctly remember, I used the address 192.168.117.129 (which
is the static address I gave to the mac) because simply
> using jackd -R -d net (which takes the default) did not work.

I'll have to try myself later.  My machines have multiple network
interfaces, which makes multicast setup a little tricky, I have to
remember how to tell the kernel which interface should be listening for
multicast, might take me a while to get that working properly.  But a
little worrying that the documentation does not seem to match the behavior
of the software.


> Master: (mac OSX Sierra 10.12.4)
>
> sh-3.2# jackd -R -d coreaudio -Cno -Pyes

Note that I'm not sure that the -C and -P arguments actually take a yes or
no value.  There were no error messages complaining, so maybe that is OK. 
The documentation seems vague and I don't have a mac to verify the
coreaudio driver behavior.  See below...

> jackdmp 1.9.11
> Copyright 2001-2005 Paul Davis and others.
> Copyright 2004-2016 Grame.
> jackdmp comes with ABSOLUTELY NO WARRANTY
> This is free software, and you are welcome to redistribute it
> under certain conditions; see the file COPYING for details
> JACK server starting in realtime mode with priority 10
> self-connect-mode is "Don't restrict self connect requests"
> Separated input = 'Built-in Microphone'
> Separated output = 'Built-in Output'
> Input channel = 0 ==> JACK input port = 0
> Input channel = 1 ==> JACK input port = 1

That seems to indicate that there are two input channels, so the -Cno did
not seem to do anything.  Maybe just leave off the -C altogether and have
only -P.
Since you have only mentioned playing audio and not capturing audio, the
goal was to get the master configured for playback only to simplify the
configuration as much as possible.

> JACK output port = 0 ==> output channel = 0
> JACK output port = 1 ==> output channel = 1
> CoreAudio driver is running...
> Starting Jack NetManager
> Listening on '225.3.19.154:19000'
> Sending parameters to ...

> Can't set net buffer sizes : Invalid argument
> SetParams error...
> Can't init new NetMaster...

If I am interpreting that correctly, the message indicates that the
netjack driver could not start correctly.

> Slave: (debian linux odroid board)
> Initializing connection with ucas-MacBook-Pro-9.local...

So the odroid was able to detect that the MacBook was running netjack.
Possibly all the later error messages are because the MacBook is not
giving reasonable responses due to whatever caused the earlier messages
about not being able to set the net buffer sizes.

> Any suggestion?

The walk-through guide says this form will print more information:
 jack_load netmanager -i "-h"

Try that and see if there is useful information printed about the number
of channels which can be used, or the optimum period size for your system.

My suspicion after looking more closely is that this message:
> Can't set net buffer sizes : Invalid argument

Is that the audio parameters set force the netjack driver into attempting
to use invalid settings.  That message comes from this location in
JackNetInterface.cpp:

        try {

            // audio net buffers
            if (fParams.fSendAudioChannels > 0) {
                fNetAudioCaptureBuffer =
AudioBufferFactory(fParams.fSendAudioChannels, fTxData);
                assert(fNetAudioCaptureBuffer);
            }

            if (fParams.fReturnAudioChannels > 0) {
                fNetAudioPlaybackBuffer =
AudioBufferFactory(fParams.fReturnAudioChannels, fRxData);
                assert(fNetAudioPlaybackBuffer);
            }

        } catch (exception&) {
            jack_error("NetAudioBuffer on master allocation error...");
return false;
        }

        // set the new buffer size
        if (SetNetBufferSize() == SOCKET_ERROR) {
            jack_error("Can't set net buffer sizes : %s",
StrError(NET_ERROR_CODE));
            goto error;
        }

        return true;

    error:
        FreeNetworkBuffers();
        return false;
    }


So it appears that the jack side was able to set the buffer size with no
exception, but then after attempting to use that information to set the
network buffer size the network API gave an error.

I just looked back at your first email and that message was not present. 
In the first configurations you explicity gave the -p128 -n2 parameters,
you did not show that in your most recent command output.  Perhaps you
introduced a new problem.
Typically when jackd starts it prints the sample rate, the sample format
(number of bits used), and the number of samples per frame and frames per
period.  I do not see  that in the most recent output you showed.
Did you truncate any output from the jackd messages, or did jackd really
not print out the buffer size information?

-- 
Chris Caudle