[Kirby]

For beginners, this is definitely good advice. Particularly the first point - if you're using a relational database, and don't structure your data around its strengths, you'll take a profound performance hit. And sure, don't pass around data structures if you don't know what you're doing, and MySQL is a pretty crummy place to put them.

However, once you're doing real work, sometimes translating to XML and back is extraordinarily expensive.

The best approach is a hybrid. Use the database to store things relationally if possible, and using defined APIs for sure. And if your translation stage is expensive, use something like memcached to make sure you do that translation as infrequently as possible. _This_ is the layer that it's appropriate to store serialized data structures at. It's not permanent storage, you can blow it away when you change the internal structures, and nobody external is relying on it. But you end up, in most programs, with even more speed benefits than if you'd stored it in the database like this to begin with - the initial build can be expensive, but then you're reading basically from memory. (Not all data is well suited to this approach, but if your data is read frequently and written to infrequently - there's few things you can do to increase performance more than this.)