Two other reasons besides mod_php quickly come to mind on why I use prefork apache–mod_perl and mod_python. The solution of using NGINX as a reverse proxy is gaining more and more popularity because NGINX is written with this in mind as a primary operating task. Apache can be used as a very effective reverse proxy as well. In fact, my first professional introduction into Apache optimization had once instance of Apache serving static content and two other instances of Apache serving mod_perl content. Regardless of how you separate things, the value is the separation and optimizing the area that will provide the greatest improvement for the effort.

One of the reasons that there are so few published benchmarks is the variety of architectures that can be used. Apache, lighthttpd, NGINX, FastCGI, Catalyst, Perlbal…the list keeps going. Combine that with custom webserver frameworks–think Python web framework–and it should be clear that there is no single solution.

The other reason is the challenge of testing real-world web application performance. It is much the same reason why you do not see worthwhile benchmarks comparing Apache to IIS. Nearly very large web application is a custom application and architecture. Building a valuable benchmark requires three elements: a testing environment that can produce a significant fraction of the production environment capacity, a testing framework that is able to emulate real-world web traffic, and a metric gathering framework that is able to measure capacity usage on a per-request basis. I have only had access to all of those elements once–and only because my team received a new production environment a month ahead of schedule.

Do not let a lack of benchmarks stop you from experimenting! Experimenting with tools such as FastCGI will give you one more tool to work with and will definitely give you a perspective that is different from the standard Apache CGI environment.

Thanks for a good article. I realize most people use apache prefork because of thread safety issues with worker (or more likely because that’s how things get set up by default out of the distro’s repos). The most common solution is yours apache for php and nginx for static files. What about using apache w/ fastcgi and having apache serve the static files as worker w/o all the overhead of mod_php for servign a gif?

I’ve been trying to find some decent benchmarks for this, but no one seems to be using worker so no one seems to be benchmarking it. I wish I had the hardware to do it myself but I don’t so I was wondering if you have considered this?

Thanks.