I am a bit worried about the performance of the new ZenMagick database layer. It seems to work reasonably fast on the demo site (I ran the trunk code last night for some tests), but it doesn’t seem to scale well with less performant systems.

Some difference in the numbers is easy to explain, though. The stats for the new code include the data to object conversion, which involves quite a bit of reflection (method_exists() for example).

However, the lower spec the system, the worse gets the ratio of using $db compared with ZMDatabase

I guess there is not too much I can do right now. The OO mapping code is the same that is used right now in lots of the service classes. Just, with migration code to using ZMDatabase it means there are more places using it.

On the other hand I have started working on reducing database access, so that might make up for some of the additional processing time. In particular products with lots of attributes should benefit from that!