LinuxMusicians

Just as a matter for comparison, the YoshimiPi (Pi4) uses ALSA directly for both MIDI input and audio output with a buffer size of 128 frames and using a cheap and simple 16bit USB/audio device. It gives dramatically better performance like this than any other combination - and I've tried quite a few.

With Rosegarden running a fairly complex 10 part session I get no Xruns at all.

I've no way of measuring the round-trip latency, but all I can say is it's unnoticeable to me

So finally got around to trying it again. And the annoying issue I have had on jack where whenever i go below 64 in buffersize i just got a constant stream of xruns. With pipewire I can now record with 32 no problem. Cool. Cool cool cool.


Btw, what jack packages can I remove after installing pipewire?

christobal wrote: ↑Mon Aug 09, 2021 3:40 pm So finally got around to trying it again. And the annoying issue I have had on jack where whenever i go below 64 in buffersize i just got a constant stream of xruns. With pipewire I can now record with 32 no problem. Cool. Cool cool cool.

Usually PipeWire uses a different period size for the ALSA device than it does for its own internal buffer. So more likely PipeWire still runs very large buffers for I/O but only calls JACK clients using smaller buffers.

With PipeWire, by default, 32 sample cycles results in a round-trip latency of 35ms. That is like running jackd with a period of around 840 (!) samples/cycle at 48k. An order of magnitude higher latency. https://gitlab.freedesktop.org/pipewire ... nt-latency has information about this.

x42 wrote: ↑Mon Aug 09, 2021 6:05 pm

christobal wrote: ↑Mon Aug 09, 2021 3:40 pm So finally got around to trying it again. And the annoying issue I have had on jack where whenever i go below 64 in buffersize i just got a constant stream of xruns. With pipewire I can now record with 32 no problem. Cool. Cool cool cool.

Usually PipeWire uses a different period size for the ALSA device than it does for its own internal buffer. So more likely PipeWire still runs very large buffers for I/O but only calls JACK clients using smaller buffers.

With PipeWire, by default, 32 sample cycles results in a round-trip latency of 35ms. That is like running jackd with a period of around 840 (!) samples/cycle at 48k. An order of magnitude higher latency. https://gitlab.freedesktop.org/pipewire ... nt-latency has information about this.

Interesting, but how does it compare with using ALSA directly - I can't seem to find any info on that.

How can i make pipewire to jack & jack to pipewire bridges?

thetotalchaos wrote: ↑Wed Sep 08, 2021 9:33 pm How can i make pipewire to jack & jack to pipewire bridges?

You don't. Just run pipewire and it acts as JACK server.