----- Original Message -----
From: "Lionel MARTIN" <[EMAIL PROTECTED]>
To: "Perrin Harkins" <[EMAIL PROTECTED]>
Cc: <modperl@perl.apache.org>
Sent: Wednesday, April 26, 2006 4:10 PM
Subject: Re: How many people use the Windows combo of Apache2 + mod_perl2 ?
Hi,
I'm replying below.
On Wed, 2006-04-26 at 17:20 +0200, Lionel MARTIN wrote:
I'm preloading this module, using "PerlModule MyLoadingModule" in
httpd.conf.
Does that change if you load them via "use MyLoadingModule" in a
startup.pl?
I can't understand how it could change using a startup.pl file? After all,
doing a PerlRequire startup.pl is just a way to incoporate code in the
main flow, isn't it?
Anyway, I gave it a try, and it gave exactly the same results:
Module Loaded: restart_count: 1 # Perl Interpreter:0x26ebbc
Module Loaded: restart_count: 2 # Perl Interpreter:0x150f29c
Module Loaded: restart_count: 1 # Perl Interpreter:0x26ec54
Module Loaded: restart_count: 2 # Perl Interpreter:0x1ac6014
I really would like to understand how the WinNT MPM is working, how
memory
is shared, in order to make our coding more efficient.
The problem is that most docs about memory management under MP is abou
LAMP,
and I'm not sure the same rules apply...
With Perl threads, no memory is shared. There is no advantage to
preloading when using threads. It just makes the threads take longer to
spawn.
I'm really sceptical (and probably lost) with what you're saying here.
First of all, from the MP docs itself
(http://perl.apache.org/docs/2.0/user/intro/overview.html#Threads_Support),
I can read that:
****************
Rather than create a PerlInterperter per-thread by default, mod_perl
creates a pool of interpreters. The pool mechanism helps cut down memory
usage a great deal. As already mentioned, the syntax tree is shared
between all cloned interpreters.
****************
And concerning me, I couldn't believe there's no difference between
preloading components, and loading them at runtime, in term of memory
consumption.
So, I made extensive tests, creating a simple component (a
PerlResponseHandler). I ma deocmparisons between preloading the handler,
and loading it at runtime, and then, using 8 concurrent accesses (with 8
Perl Interpreters enabled by default), and I could see some difference in
terms of memory consumption.
I have got a lot of different results, but for example:
When preloading my compoennts, and asking Apache to start 8 Perl
Interpreters and then, making 8 concurtrent requests, this was taking
27968 kb of RAM for the main process and 29596 kb of RAM for the child
process.
On the other hand, without preloading the handler, and then, making 8
concurrent requests (=> handler loaded on the fly), this was taking
11740kb for the main process and 35252kb for the child process...
I don't know how to exactly interpret that, but that shows there are some
differences.
I really would love to understand how sharing is working, and what's
exactly shared (or perhaps have the confirmation that nothing is shared
under winnt MPMs...?)
I think that I should better try with LAMP as well, but to be honest, I
have never tried to install Linux.
Have you got any resources where I could read about how things are managed
under Windows?
- Perrin
