The short answer: #include <jack/jack.h> #include <jack/ringbuffer.h> #include <semaphore.h> sem_t block_wait; jack_ringbuffer_t ringbuf; init() { int max_blocks = buffer_size_in_samples/jack_block_size sem_init(&block_wait, 0, max_blocks); } jack_process() { float *out = jack_port_get_buffer(...); jack_ringbuffer_read(ringbuf, out, block_size*sizeof(float)); sem_post(&block_wait); } int exit = 0; worker_thread() { while( !exit) { sem_wait(&block_wait); read_from_disk_to_ring_buffer() } } correct program should use jack_default_audio_sample_t instead of float and should deal with buffer size changes (as well as sample rate) ;) I'll try to write a working example tonight (using libsndfile of course). Or take a look at examples/capture_client.c in jackd distribution. It uses mutex/cond but the idea is the same (semaphores are better, actually. Jack example needs a fix). Regards, Dmitry.

Hello, ok, read the chapter semaphores! B)) Am Freitag 14 Juli 2006 12:17 schrieb Dmitry Baikov: fine , I understand! Just to be sure. This means every Port has his own Ringbuffer, right? So stereo needs two, 5.1 needs 6 ? But i only need one sem_post for all, because i should avoid to have different readpositions in my RingBuffers. And if i understand Paul Davis in the right way, i never read more data as jack needed, and this isn't more as 1 or 2k - normally. ??! sizu c~

On Fri, 2006-07-14 at 13:27 +0200, conrad berhörster wrote: no, the workerthread should *not* be called more often than the jackthread. either the same frequency or slower. it should typically end up processing more data than the jack thread does per call. from experiments done in about 2000, if you must absolutely avoid situations where one thread outruns the other, linux tends to need about a 5 second buffer if the worker thread is not running realtime (which it probably should not do). it also seemed that we got the most efficient disk i/o rates when writing chunks of about 256kB to disk rather than larger or smaller bits.