http://bulk.fefe.de/

An informative, and fairly technical, read on network and filesystem scalability bechmarks. The first 2-links are Unix-only (*BSD and Linux), the last link includes various Windows and Solaris versions.

What comes out on top? The answer won't surprise you, neither will the identity of the worst-performing OS. What is shocking is how badly some otherwise decent OSes perform. The section on filesystems is also quite interesting: you thought non-journalled filesytems were faster than journalled ones? Think again.

Quick FAQ on the article: O(n) means whatever is being measured proportionally increases as n increases, O(1) means performance is constant no matter how big n is. For instance: if you were measuring the amount of time it takes a webserver to serve requests, an O(n) performance means the time the webserver needs to serve a request increases proportionally to the number of requests; an O(1) performance means it takes the same amount of time for the webserver to serve 1 request as it would a gazillion bazillion requests.

saidi