A master process creates a pool of N processes. Tasks can be submitted
asynchronously as a function
f and its argument
x. As soon as one of
the processes is available, it computes
f x and returns the result.
This library allows to take advantage of multicore architectures by message-passing and without blocking. Its implementation relies on fork, pipes, Marshal and Lwt.
- interface may still be subject to slight changes;
- passed a few units tests;
- used stream interface successfully at full scale.
- observed 5x speedup on 8 cores when converting a stream of lines from one file to another. A task consisted in parsing a line, converting the record, doing one in-RAM database lookup per record, and printing the new record. Throughput was 50K records per second, using a granularity of 100 records per task.
Do not hesitate to submit experience reports, either good or bad, and interface suggestions before it is too late.