The "crucial" part of vmalloc is as I see mmap. I do not get any significant reduction in overall memory footprint when many threads are created/exited. For example, I start 150 threads and the memory footprint goes to about 380-420 MB (Mac OSX). Then I stop all those threads and I get down to about 320-330 MB. I believe the effect of vmalloc is not really visible unless you use mmap to allocate memory.
Right, if you use standard malloc, it is up to it how to return memory and it has its own rules and its own fragmentation. Still 420Mb and 330Mb is 100Mb less if that makes a difference. I would suggest to run it for a while when fragmentation gets worse, this part is not clear and we are still experimenting as well, of course with mmap-ed pages it is easier but it just proves your thought: there is not one-for-all-cases tool exists today
-- Vlad Seryakov 571 262-8608 office [EMAIL PROTECTED] http://www.crystalballinc.com/vlad/
