On Wednesday 19 September 2001 02:05 pm, Ryan Bloom wrote:
> On Wednesday 19 September 2001 01:59 pm, Jeff Trawick wrote:
> > Ryan Bloom <[EMAIL PROTECTED]> writes:
> > > On Wednesday 19 September 2001 01:21 pm, Greg Ames wrote:
> > > > For now, ThreadsPerChild might be fine. I'd feel a little better
> > > > about some small constant, like 5 or 10. But the code does need to
> > > > learn to deal with "queue full" more gracefully.
> > >
> > > Guys, please take a look at the archives. Manoj did these tests over a
> > > year ago. ThreadsPerChild is the optimal value. We have run the
> > > numbers once already.
> >
> > I would think that tests for reasonable queue limit need to be done
> > with an MPM hich is otherwise working fairly reasonably (e.g., not
> > losing connections when the queue fills). Surely we haven't been
> > spinning our wheels all this time when Manoj had one lying around :)
>
> The tests were done before the MPM logic was created. And yes, the worker
> MPM does go back to the very first design that Manoj and I wrote in 6
> months while we were both back at IBM. Why do you think I keep saying that
> we are constantly re-visiting decisions that were made in the past for good
> reasons.
>
> The worker MPM is not new, the fdqueue logic is not new. The design comes
> from the Lotus Domino Go WebServer, and the code was written over three
> years ago.
Take a look at:
http://cvs.apache.org/viewcvs.cgi/apache-apr/pthreads/src/main/fdqueue.c
specifically at revision 1.12, where I committed code to have
slots in queue == Threads per child.
This would be the time frame to review the archives.
Ryan
______________________________________________________________
Ryan Bloom [EMAIL PROTECTED]
Covalent Technologies [EMAIL PROTECTED]
--------------------------------------------------------------
