Message-Id: <199701081618.LAA19189@postman.osf.org>
To: new-httpd@hyperreal.com
Subject: Fwd: mod_perl's performance
Date: Wed, 08 Jan 1997 11:18:18 -0500
From: Doug MacEachern <dougm@osf.org>
Sender: new-httpd-owner@apache.org
Precedence: bulk
Reply-To: new-httpd@hyperreal.com

Since mod_perl has come up a bit recently and I have been asked about
mod_perl vs. CGI performance, I though some might be interested in
this recent message.  Keep in mind, the script he's benchmarking is
_very_ small, so this does not illustrate the speed up mod_perl gains
by avoiding Perl startup-time.

-Doug

------- Forwarded Message

From: palowoda@fiver.sns.com (Bob Palowoda)
Message-Id: <199701070816.AAA04719@sns.com>
Subject: mod_perl's performance
To: dougm@osf.org
Date: Tue, 7 Jan 1997 00:16:01 -0800 (PST)
X-Mailer: ELM [version 2.5 PL0a3]
Content-Type: text
X-UIDL: 6012ee5d47a4811782cbedc7dea4d498


Doug,

Although I am not an active participant of the development
of mod_perl I think you may find this interesting.  First
let me introduce myself.  I work at Sun Microsystems as
a Web Server Perfromance Engineer.  Usally what I do is
measure and identify components that have a significant 
impact on web servers.  Recently I became interested in
the performance of mod_perl.  I just got done doing 
a Webstone 2.0 run on the differences between using
a 100 percent cgi load via fork/prefetch/parse and 
mod_perl. 

The testbed consists of 1 Ultra200 1Gmem (2 200mhz cpu's) with
six Ultra200 acting as loads on three 100mbit networks. 
The switch is a Synoptics 5000 with a 10G routing backplane,
ATM FDDI etc.  In this test case we only use one 100mbit
net because the loads of a simple cgi script the prints 
the environmental variables is quite substantial with
only 50 load clients.  What happens during the tests is
50 clients are spawned on each load machine and simply 
excute the printenv for the spawn perl and registry.fpl
for mod_perl.  Variable count return is the same for
both scripts. 

At a load of 50 clients for a two minute test run   
Apache 1.2b4 and the forked perl we have:

Perl forked and excute script:   66 cgi/ops/sec

Now with Apache1.2b4 and mod_perl-0.90 we have:

Perl embedded                :  246 cgi/ops/sec


This is an extremely good performance increase of 
what approximately 400 pecent for perl cgi execution.
Further mod_cgi decease the load fact by nearly 4 times.
Compared with with other methods which I'm not a liberty
to discuss I would say your work and the others that
have contributed will have an impact on the web.  Remember
this was a crude parliamentary test.  In fact I have to
increase the load on the mod_perl for better results. 
It's a relative comparison anyway.

You may pass pass the information along to the mailing
list if you like.  I hope the group continues on for
a little while as I believe alot of developers will
be using mod_perl. 

Regards,

Bob Palowoda


- -- 
+--------------------------------------------------------+
| palowoda@fiver.sns.com http://fiver.sns.com/~palowoda/ |
| Solaris x86 Corner http://fiver.sns.com/               | 
+--------------------------------------------------------+


------- End of Forwarded Message