The plots below show the output of a simple program measuring the time between process callbacks for various values of the period size (P) and number of periods (N). X-axis = periods, Y-axis = milliseconds.

The program doesn't use Jack but interfaces to the ALSA device directly using libzita-alsa-pcmi (I will include it as one of the example applications in the next release).

The sound card used is an RME Babyface Pro, with a sample rate of 48000. Jack and some clients are running at the same time, using a PCI bus based card.

Times between period starts seem to be clustered around multiples of 1ms as could be expected for a USB device.

In the P = 96, N = 3 case, a difference of 3 ms occurs approximately every 10000 periods, which means that the sample frequency is around 50 ppm slow compared to the 1 ms clock used by the USB subsystem. Strangely there are no occurences of 2 ms for the P = 48 case.