A copnstraint or a coincidence. But I would tend to forgive an optimizer
that didn't keep track of the coincidences, since keeping track of them
would imlpy a fairly large cost for fairly low gain.
BTW, in RDB Rdb the DB builder can specify placement in such a way that the
material required for the join can be in a single database block. That
means that the overhead of actually performing the join is CPU overhead,
but not disk I/O overhead. Generally speaking, CPU overhead is so low as
to not be worth considering in cost based optimization.
Received on Mon May 03 2004 - 14:22:58 CDT
