ALSA newbie warning...

I have a funny application that does av sync by using the timing of a callback to set the audio timebase. In other words, I need to invoke that callback "as close as possible" to the time a previously-sent buffer of samples is presented, i.e. exits the speakers. The application uses the clock time of the arrival of this callback and does some math to figure out whether to repeat or skip video frames. I do not like this arrangement, but unfortunately I am stuck with it. My job is to hook this app to ALSA audio output.

My current design uses a producer-consumer model with a thread that writes fixed-size application buffers to the ALSA ring buffer, driven by a message queue fed by the application. This is a little redundant I think, but I do not own the buffers sent by the app so this is what I came up with. The application needs a regular callback with the timing described above each time one of their buffers is played, and it is assumed they are played in sequence.

I tried setting the ALSA callback to trigger when there are an application-buffer-size amount of frames available, using snd_pcm_sw_params_set_avail_min(). My idea was that if I primed the ring buffer to be full before it starts transfers with snd_pcm_sw_params_set_start_threshold(), I would get the ALSA callback each time that number of frames exited the ring buffer. But this did not work: I seem to be getting callbacks much more frequently that I wanted, probably on ALSA period boundaries. So this means I do not understand how snd_pcm_sw_params_set_start_threshold() is supposed to work I guess.

Will somebody please comment on the viability of and/or problems with my current design, or suggest an alternative way to configure ALSA that will give me the output timing I need?

Currently I am testing on an x86 ubuntu box, kernel 2.6.22-14, ALSA 1.0.14, gcc 4.1.3, glibc 2.6.1, but will also cross-compile for several embedded platforms in the future.

Thanks in advance.