Loadtest

I am writing a loadtester for our application, and I am testing against
IIS/Tomcat versus Apache webserver/Tomcat.
I am threading connections to the webserver by opening a socket and
writing to the outputstream.

Socket server = .....
OutputStream toServer = server.getOutputStream();
toServer.write(request);

My question is probably more on the webserver side, is this an
appropriate way to handle sending requests to a webserver,by opening
sockets? If I create a multi-threaded application and send 500
different requests, is this the same as 500 client browser requests, as
far as the webserver is concerned?

- Eric

ericwaite@gmail.com napisa³(a):
> I am writing a loadtester for our application, and I am testing against
> IIS/Tomcat versus Apache webserver/Tomcat.
> I am threading connections to the webserver by opening a socket and
> writing to the outputstream.
>
> Socket server = .....
> OutputStream toServer = server.getOutputStream();
> toServer.write(request);
>
> My question is probably more on the webserver side, is this an
> appropriate way to handle sending requests to a webserver,by opening
> sockets? If I create a multi-threaded application and send 500
> different requests, is this the same as 500 client browser requests, as
> far as the webserver is concerned?

Hello,

there are many possibilities to do that. Let's look at Apache Benchmark
tool ("ab" binary in Apache's bin/ directory,
http://httpd.apache.org/docs/2.0/programs/ab.html):

* it can "employ" multiple threads ("-c" option) simulating a number of
client browsers accessing server resources at one time instead of
sending sequential requests
* it can use Keep-Alive HTTP option ("-k") to send requests without
disconnecting the socket (if server accepts it, of course)

There are also different types of threading models in Apache HTTP
Server. The mpm_prefork model uses new child processes to handle
incoming requests. So you'll probably get varying results when testing
threaded and non-threaded servers.

For example, the new experimental (used in Apache 2.2 only)
mpm_event_module is said to handle incomming connections better by
keeping helper-threads to do some initalization work instead of
taking listening (main) thread's time.

And yes, creating new threds to open sockets will cause the server to do
same things as it was real browsers. Although, to better simulate
real-world's browsers your tool should use Keep-Alive option sending
multiple requests (1 thread == N requests, N>1) until server sends
"Connection: Close" header in response (AFAIK it happens, when too much
clients want to have alive connection).

So, to simulate 500 real-world browsers, create 500 threads sending N
requests each. N can be random number in some range. IMO, that's the
best way to simulate heavy traffic on production server..

--
Pozdrawiam / Best regards
Pawel Zdziarski
gmail: faxepl