Re: New FreeBSD port ALSA to JACK MIDI, a2jmidid

Ok, finally got around to measure and compare MIDI latency.
The setup was:

jack_midi_latency_test -> JACK -> jack_umidi -> USB -> MIDI out -> loop cable 
-> MIDI in -> USB -> jack_umidi -> JACK -> jack_midi_latency_test

... and the same with alsa-seq-server and a2jmidid instead of jack_umidi.

Findings:
 - JACK processes MIDI in the same cycles (period) as audio.
 - The smaller the JACK period setting, the lower the latency.
 - Latency correction values set by JACK OSS backend are ignored.
 - alsa-seq-server and a2jmidid add noticeable latency vs. jack_umidi.


For example a period of 384@48kHz (8ms) with jack_umidi reports

Reported out-port latency: 0.00-0.00 ms (0-0 frames)
Reported in-port latency: 0.00-0.00 ms (0-0 frames)
Average latency: 11.93 ms (572.41 frames)
Lowest latency: 8.02 ms (385 frames)
Highest latency: 16.00 ms (768 frames)
Peak MIDI jitter: 7.98 ms (383 frames)
Average MIDI jitter: 1.99 ms (95.13 frames)

whereas with alsa-seq-server and a2jmidid it is

Reported out-port latency: 0.00-0.00 ms (0-0 frames)
Reported in-port latency: 0.00-0.00 ms (0-0 frames)
Average latency: 18.93 ms (908.83 frames)
Lowest latency: 18.04 ms (866 frames)
Highest latency: 21.81 ms (1047 frames)
Peak MIDI jitter: 3.77 ms (181 frames)
Average MIDI jitter: 0.29 ms (13.50 frames)

which means it adds 7ms roundtrip latency on average.
For 192@48kHz (4ms) we get

Reported out-port latency: 0.00-0.00 ms (0-0 frames)
Reported in-port latency: 0.00-0.00 ms (0-0 frames)
Average latency: 5.97 ms (286.44 frames)
Lowest latency: 4.02 ms (193 frames)
Highest latency: 8.00 ms (384 frames)
Peak MIDI jitter: 3.98 ms (191 frames)
Average MIDI jitter: 1.00 ms (47.32 frames)

versus

Reported out-port latency: 0.00-0.00 ms (0-0 frames)
Reported in-port latency: 0.00-0.00 ms (0-0 frames)
Average latency: 10.78 ms (517.51 frames)
Lowest latency: 10.00 ms (480 frames)
Highest latency: 13.46 ms (646 frames)
Peak MIDI jitter: 3.46 ms (166 frames)
Average MIDI jitter: 0.27 ms (12.53 frames)

which is a plus of ~5ms roundtrip latency on average.

Please note that using jack_umidi, the latency values are spread uniformly 
between 1 and 2 periods, whereas alsa-seq-server produces more like a gaussian 
cut at the lowest latency.

Conclusions:
 - For live playing, jack_umidi and small JACK period are preferable.
 - We have no way yet to adjust the MIDI latency to audio latency.
 - "Reported out-port latency" suggests possible latency correction values.
 - It may still be worth looking into the JACK ALSA MIDI backend for this.

@HPS: Any ideas about the latency introduced by alsa-seq-server / CUSE?

@Anyone else: What are your MIDI use cases where latency matters?

Regards,

Florian