There are two types of people when it comes to server performance: There are the benchmark/log/process manager people who studiously scrutinize everything from how often a machine to pages to the load per core on a machine, and there are the more laid-back folks who use the rather practical and somewhat easier benchmark of “If it hasn’t started to show a performance hit, don’t worry about it”. If you’ve been a frequent reader of the blog, you may be expecting my tried and true “Well… you should be somewhere in the middle…”
Surprise! I side with the practical guys on this one. I’ve heard the arguments for obsessive monitoring of a server to gauge its performance, but such monitoring should be a secondary measure. I don’t intend to say that it’s wrong, because it isn’t; process and resource management is an important thing no matter how you slice it, and you should always have someone or something looking out for it. When it comes to server performance, however, setting arbitrary benchmarks as markers to your server’s performance isn’t exactly the most useful thing to be doing with your time. In a production environment, the biggest concerns you should have are stability and speed, in that order; if a server’s performance has been and continues to be stable and fast, just keep an eye on it and let it alone. Don’t obsess over reducing the load per core by .5 or whatever you might think is a proper performance benchmark; as long as there’s no noticeable slowdown, you’ve done a good job so far (For, as we know, a server admin’s job is never done…)