Mailing-List: contact new-httpd-help@apache.org; run by ezmlm
Precedence: bulk
Reply-To: new-httpd@apache.org
Sender: gregames@Mail.MeepZor.Com
Message-ID: <3AF1B706.9E86E837@remulak.net>
Date: Thu, 03 May 2001 15:52:38 -0400
From: Greg Ames <gregames@remulak.net>
MIME-Version: 1.0
To: new-httpd@apache.org
Subject: Re: cvs commit: httpd-2.0/server/mpm/threaded threaded.c
References: <Pine.LNX.4.30.0105030845160.22008-100000@koj.rkbloom.net>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit

rbb@covalent.net wrote:
> 

> >
> > Wait a second, our user configured the server so that the child process
> > never terminates.  That's what our docs say that directive does.  This
> > says, well, we just take a really long time to terminate.  Either the
> > patch needs to be backed out, or the docs need to change.  I personally
> > would rather keep the 0 meaning the child process never dies.
> 
> I should clarify quickly, I have stated a preference, but I will be
> perfectly happy as long as either the docs change or the code does.
> 

Actually, I thought about it a sec myself, and maybe I should go update
the docs after all (if I can find them all :-).  Here's why:  

In a previous life, I was a performance dude for OS/390's TCP/IP stack. 
We set a specWeb96 record that blew away the previous record by a large
margin.  It sustained about 21,000 requests/sec.  (OK, I'll admit we
cheated by using a 10-way IBM S/390 G5 mainframe, and we used AFPA
caching, and we tweaked the begeezus out of the TCP/IP stack and the
Domino Go threaded web server, but big deal...)  That was maybe 3 years
ago, and the record has been beaten many times since.

So if we just take 21K/sec, assume it's sustained for 24 hours, and do a
little math, I come up with 1.8G requests per day.  It's true that the
vast majority of Apache users won't have a honkin' 10-way SMP box at
their disposal, and we're just talking one process here on a web server
that just barely became threaded and still does a lot of mallocs and
mutexes.  But Moore's Law still applies, and we know how to fix
scalability issues, so yeah, it's actually possible that an admin could
notice this change some day.  I truly hope they do :-)

I'll update the docs.  But first I need to get a trap ready for the
vhost overlay seg faults on daedalus.  We're seeing them about once
every 2-3 days now.

Greg