At the moment I am developing a multi-threaded compressor on Haiku-OS.
How despise the machine being a I7 with four real CPUs and four virtual CPUs I
don't see a real speedup once I configure to run with more than four >threads.
Between one to four threads i see a real speedup of the compression, but after
that the speedup slows down a lot.
Since I am running the threads at NORMAL_PRIORITY I was expecting the
slow-down to take place after I had eight threads running.
However, I am running the each thread to wait on it's command queue using a
snooze(1LL) inside the loop. Even then I am using a lot of CPU time >in the
loops. Add in the cache misses because different threads are accessing
different memory areas and I am not surprised that the threads use a >lot of
CPU while idling.
What I wanted to asked before doing a major rewrite, does anyone know if
resume_thread() and suspend_thread() are very fast calls?
