At 04:12 PM 7/23/02 +0200, Arthur Bergman wrote: >>Well, the software we had on 5005 seemed to work, but now any thread >>calls within a thread are executed in series. This is, it seems, why my >>crazy idea for a Thread manager didn't work as the management thread >>couldn't spawn any more processes. I am not 100% that this is the >>problem, but it would explain a few things. >This should work, there are even testcases trying it out.
Hmmm.. are they testing serialization as well? I just realized that the multi-thread job submission test in Thread::Pool just tests whether the jobs are handled in the right order, but not whether they're really handled in parallel. On the other hand, some tests would take a _lot_ longer than they would when done in parallel, so I guess it does parallellize properly... >You are sure you are not holding some lock that causes them to be serialized? In the test case, I do a sleep( rand(3) ) inside the threads, to have some randomness in the order in which the threads are handled. If you could do that as well, do you still see the same serialization? Liz
