Re: [tlug] Apache MPM performance

["Nguyen Vu Hung" <vuhung16plus+shape@example.com>]

2008/3/12, Curt Sampson <cjs@example.com>:
> On 2008-03-11 21:57 +0900 (Tue), Dimitar Dimitrov wrote:
>
>  > In your case, as you are using ab, which is a single-threaded
>  > benchmark you are not testing any of the advantages of the MDM module
>  > or the machine you bought.

> Nguyen, you're using ab? There's your problem right there; It sends a
>  single request, waits for a response, and loops. It's fine for showing
>  the minimum speed of your web server, but not useful for determining a
>  reasonable maximum, or seeing if your server will fall down under load.
>
Yes, I am using ab, which can be use for DoS. What is wrong??? :)

My system is quite unique. At first I have an .so file which is
written in C/C++,
then a PHP extension built with it.Each time I call my php extension's API,
the C/C++ code is excuted. By the nature of the .so and the extension, time
for a PHP call is 1.0 - 4.0 seconds.

So if I run

./ab -n n -c c

or

httperf --num-conns=c --num-calls=n

PHP module has enough time spawning httpd proccesses ( prefork or worker ).

I've run both ./ab and httperf against 2 servers, and they gave the same result.

I don't know if is there anything with my test scenario, but the
results look good.

>  What you want is httperf, which will send a fixed number of requests per
>  second and tell you how many actually get a response and give you some
>  stats on response times. Basically, you just keep running it with a higher
>  and higher number of requests per second until your web server collapses
>  and dies, and then you have a better idea of what load you can handle.
>
See above. But I start with a large number of connections ( 2000 ).

>  You probably want it running on a separate machine from the web server
>  for serious benchmarking. However, it's fortunately single-threaded
>  (event-driven), so while it will eat up all the CPU it can, it won't use
>  more than one core, so on a multi-core machine it ought still to be a
>  reasonable benchmark.
I run the rest from a Xeon 3.2GHz, 2 CPUS, 8 cores, 8GB of RAM, 1000MB
ethernet against Solaris T5220.
Is there something wrong with it?