Justin Erenkrantz wrote:
[ Moving this part of the discussion to [EMAIL PROTECTED] ]
On Fri, Nov 23, 2001 at 03:16:35PM -0800, Brian Pane wrote:
Thanks. Here's my patch to optimize away the mutex operations in pools that have been designated thread-private. With the current worker code, it can eliminate the mutex ops for subrequest pool creation/destruction. By combining it with your worker redesign, I think we may be able to eliminate the mutexes for the ptrans pool.
Nice idea. However, I think it might be best to resurrect
Sander's patch which did a rewrite of the pool code (keeping
virtually the same API) but made the internal code much cleaner. And, it had thread-private to boot. (The current pool code has
threads grafted on to it, so it isn't as clean.) And, we proved
it was a performance win, too, I think.
Agreed--now might be a good time to revive Sander's pool patch.
The benchmark results for my pool patch show that:
* Modifying the pool code to reduce mutex contention is a good idea. With the patch applied, we saw a large decrease in run queue length and a modest increase in throughput on Solaris.
* My patch isn't ready for production use. There's an apparent race condition in my patch that results in rare heap corruption, and so far I've been unable to isolate the bug.
Thus I think Sander's patch may be a better choice than mine.
--Brian
