Up and Running

This domain is being served from a slicehost.com VPS account, which I’m sharing with my friend Danny. We spent a few hours installing a typical LAMP environment from Debian packages. It all went smoothly; we’re pretty much set up. Eventually, we’ll get around to putting actual content up on our respective domains, but we expect traffic and load to remain very light. So we spent most of our time dealing with the main resource limit on our “box”: memory.

There’s lots of information available on various optimizing strategies for Apache and MySQL, but it basically boiled down to 2 things for us. In Apache, we lowered the number of start/max/min servers, and loaded only modules we needed. In MySQL, we decreased the sizes of various buffers and queues, and also disabled InnoDB support. The latter alone changed the initial size of MySQL from 20 megs of resident memory to 9 megs! I’m sure we’ll continue to tweak later on, but memory usage seems fairly reasonable right now for what we’re running.

It’s funny to reflect on how far open source software has come in general. I remember talking nine years ago with some corporate guy in New York about a piece of open source he promptly dismissed as “not ready for the enterprise” (that phrase always makes me laugh. Whose enterprise, exactly? Mine?). Back then, it was common to hear disparaging stuff like that all the time.

And now–is there anything that’s NOT enterprise ready?! Maybe the suits are happy, but I’m not sure everyone else has benefitted. As the software has grown in scalability, performance, and features, the requirements have also increased drastically. And that means the bar is higher for even a fairly standard LAMP environment, for example. Default installs of Apache, MySQL and PHP can be real memory hogs.

I’ve been discovering a few interesting alternative projects with lighter feature sets. lighttpd and thttpd are both high-performance, small-foot web servers. And a lot of folks seem to be looking to SQLite if they don’t need the fancier features of a complex RDBMS. Maybe I will experiment with these at some point.