When does a merge join beat a hash join?

A general (theoretical? academical?) question:

When the Oracle optimizer chooses a merge join, why doesn't it choose a hash join instead?
Shouldn't a hash join always be faster than a merge join? And if it isn't, is that because of poorly chosen parameter values, or are there data situations in which it is inherent that a merge join is faster than a hash join?

Jaap. Received on Wed Mar 05 2003 - 02:58:43 CST