Collabora Keeps Pushing PulseAudio For Android

[raindog469]

And what are you going to do with JACK support on an Android mobile device?

Record music while listening to what's already been recorded, do live effects ... actually, that's all I want, those two things. On my tablet, not my phone.

As was once boldly and plainly declared to me by one of the JACK maintainers on IRC, "it is utterly impossible to do 3d graphics and zero latency audio on the same system at the same time, so if you are trying to do that, you should give up now."

Good thing we don't need zero latency, then. There's a pretty huge gap between zero latency and the 50-100ms range of a typical PulseAudio setup. I'm happy with 5ms, and if Pulse can do that on the Android hardware that's out there, well, great.

you pretty much can't mix most drivers with an appreciable real-time system like JACK, where a dropout means "game over" (the sound server literally dies if it drops out at all).

Even with a "realtime" kernel, Linux is not hard realtime and neither is JACK. On my system, a dropout means the "xruns" count in qjackctl increments and I mutter obscenities and remember the timestamp so I can go back and check for clicks when I'm done. No server death unless you want it.

[liam]

Good thing we don't need zero latency, then. There's a pretty huge gap between zero latency and the 50-100ms range of a typical PulseAudio setup. I'm happy with 5ms, and if Pulse can do that on the Android hardware that's out there, well, great.

Arun demonstrated that you can easily get 20ms with PA on Android. Now considering even the iphone ends up with around 50ms or so latency, the extra 15ms that PA currently adds doesn't seem like much. When he finishes the port, we might be able to expect even less latency.

Even with a "realtime" kernel, Linux is not hard realtime and neither is JACK. On my system, a dropout means the "xruns" count in qjackctl increments and I mutter obscenities and remember the timestamp so I can go back and check for clicks when I'm done. No server death unless you want it.

The problem isn't that Linux can't do hard rt or not but rather the particulars of the system. RTOS' are typically made with very particular constraints in mind and a desktop is pretty much at the opposite end of simple controllers you seem them on.
An interesting place to see read about various systems latencies is OSDAL. They'll run their systems for a year at a time and chart the latencies. Even under load, the jitter is surprisingly low on these general purpose machines.

[curaga]

and by not using Jack unless you actually need REALLY low latency, you still get PA's many advantages including simd ops.

Look ma, my infinite loop is SSE4 optimized!!111

[Lynxeye]

PulseAudio can achieve very low latencies. The only thing is: your apps need to communicate their latency needs to the Pulse server, otherwise it tries to save energy and work with a higher latency and only adjusts to your apps need after there has been drop outs.

So when app writers finally accept Pulse as the feature-rich audio system of the future and build their apps nativly on top of Pulse and not some ALSA-Pulse plugin, I will be easy going for them to meet their latency needs.