GA and performance concerns Re: 2.0.23 tarballs up

[Brian Pane <bp...@pacbell.net>]

Bill Stoddard wrote:

>+1 for beta.  -1 for GA.
>
>Our announcement letter for the beta should clearly state that the server API is stable
>and third party module authors can begin porting their modules to this code base and that
>the beta is a candidate for GA. If this beta holds up in the field for a few weeks, then
>declare it golden. Cliff's comments regarding known problems should go into the
>announcement letter too.
>
>We've been running apache.org for months with fewer and fewer problems with each rev. A
>number of other folks are sucessfully running httpd 2.0 on production sites. I believe the
>prefork beta -will- hold up. One thing that concerns me... Do we have a good feel for the
>performance and memory differences between prefork 2.0 and 1.3?  It would really stink to
>discover that prefork 2.0 is way slower or has a much bigger memory footprint than 1.3.
>Some capacity planning info would be good to have...
>
I don't have 1.3 vs 2.0 direct comparisons, but what I do know from
performance profiling of 2.0 is that the filters code is consuming
a lot of CPU in 2.0, especially for mod_include (which spends a
surprising amount of time splitting buckets and brigades).

I propose that some benchmarking vs 1.3 with mod_include (plus any
other brigade-intensive modules that others might be able to identify
as good test cases) should be a prerequisite for 2.0 GA.  (Hypothesis:
2.0 prefork is faster than 1.3 for static, nonparsed file delivery
due to its use of sendfile, but it's slower than 1.3 for parsed or
dynamic content.)

--Brian