I ran a show with QLab and hosted a live vocal effect on the same Mac. Here was my setup:

What's odd is that Pd would have ~300mS of latency if my effect patch was started after QLab, but only about 10mS if it was started before. This morning I finally upgraded to Pd 0.55.2 from 0.55.1 and tested with the latest version of QLab, and latency is still dependent on startup order, although only about half as bad (~150mS) in the worst case (and also ~10mS in the best case). Pd's audio settings are 44100, 3 mS delay, 64 sample buffer size, and not using callbacks.

Does anyone know why Pd's latency is dependent on the order it is started? Are there other tests I should perform to understand better what's going on? This behavior is somewhat surprising to me because 2 years ago I ran a much more complicated show involving live effects implemented with REAPER, plugdata, VSTs, and OSC--I don't remember there being any issues.

@jameslo Good soundcard the Ultralite Mk4..... I bought 2 recently to replace old gear.
I can only guess that QLab has set a large buffer for the Mac communication with the soundcard (default 512).... and Pd is then having to use that for its data exchange with the sound card.... but if Pd starts first it sets the buffer size and then QLab uses Pd's smaller buffer size?
David.

@whale-av My fancy colleagues look down on the ultralight, I have no idea why. I worked in 4 different theaters that owned the mk3 and was always happy with them.

Is it true in my case that both Pd and QLab are accessing the same MOTU driver, and that that driver implements the coreaudio API, and that that API governs how applications share drivers?

Is there any way to check what buffer size the MOTU driver is using?

One of the tests I ran this morning was to start QLab, then start my Pd patch, then quit QLab and measure. Same long latency. Maybe I should try switching Pd away from the ultralight and then switching back before testing? That would test your hypothesis, wouldn't it? Better yet: run QLab, quit QLab, run Pd, then test.

@jameslo Mk3 was good, but the web interface of the Mk4 works much better than cuemix. They have gone back to cuemix with the Mk5 though, which is why I bought the Mk4.

I don't know how much info the Mac gives about the driver.
The test you ran this morning gave the result I would expect.
I believe you can set the buffer size in QLab..... so you could try to set it low.
I think the Motu also has a setting. In the past (using windows) they had to match, but maybe the mac driver is more automatic.
David.