Message-ID: <3713D2B0.9F44B7C6@raleigh.ibm.com>
Date: Tue, 13 Apr 1999 19:26:40 -0400
From: Bill Stoddard <stoddard@raleigh.ibm.com>
MIME-Version: 1.0
To: new-httpd@apache.org
Subject: Re: apache-apr
References: <Pine.LNX.3.96dg4.990413144834.24759q-100000@twinlark.arctic.org>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: new-httpd-owner@apache.org
Precedence: bulk
Reply-To: new-httpd@apache.org

Update on benchmarks. All standard disclaimers apply...

A while back I posted some results between the queued accept (listen
threads queuing work for worker threads to consume) and the
poll()/accept() model. I was really suprised to find less than a 1%
difference between the two when listening on a single socket. Since
then, a lot of things have changed in the code base. It's cleaner, my
multi-accept code is much cleaner and the results have changed. Here
they are:

All tests run using ApacheBench on AIX, 30 concurrent clients requesting
a 500 byte file using HTTP/1.0 to a single socket. (Best of three runs
recorded) Hybrid Apache (a.k.a. apache-apr) was started with a single
child process

Number of sockets	Apache 1.3.6		Hybrid Multi-Accept	Hybrid Queue
1			337 cps			349 cps			317 cps
10			332 cps			337 cps			317 cps
40			308 cps			315 cps			314 cps		

The results suggest that a site heavily using IP based multiple virtual
hosting may benefit from the queued accept model. 

I'll commit the multi-accept code later this week.

-- 
Bill Stoddard
stoddard@raleigh.ibm.com