simulate server load with BSD sockets

I'm using blocking TCP sockets in C and I want to simulate a high load on the server when there are many simultaneous connections and then I want to measure the time necessary to access the server via a browser during this high load time (the server understands HTTP headers). Also each client request ends fast (sends a HTTP header - gets text).

How do I do this (without crashing my local machine -> I tried using fork to make many clients; also, I have a virtual machine too). If anyone has an idea or some general directions about how to do this, it would mean a lot.

Edit: I need to run this with my own client, which uses a modified version of the OpenSSL library to connect to my SSL/TLS server, so I can't use external test tools.

I want to know how to开发者_JAVA百科 build the client and the server. I don't know too much about other sockets than the blocking ones, I'm just skimming through the UNIX Network Programming book of Richard Stevens, but I was wondering if anyone could point out the exact solution.

Thank you !

The easiest resolution to this would be to download an existing stress testing framework such as fwptt ( http://fwptt.sourceforge.net/ ).

If you want to implemennt your own stress testing framework, I'd suggest you lose the blocking nature of your code and go with a parallel design that will scale beautifully. The limiting factor is pretty much your CPU then.

Having two physical servers would be ideal, so that then your stress test isn't affecting the CPU (and therefore the response times) of the server. Also that VM of yours drains up precious CPU time.