On Aug 17, 2006, at 1:41 PM, Peter Eisentraut wrote:
> But we need to work this from the other end anyway. We need to
> determine first, what sort of statistics the planner could make use of.
> Then we can figure out the difficulties in collecting them.
>
There are still some things that the architect or DBA will know that
the system could never deduce.
Any suggestions for what these statistics are? Cross-column
statistics have been mentioned previously. Another that's come up
before is how "clustered" a table is around its keys (think web log,
where all the session records are going to be in the same page (or
small set of pages)). FK selectivity has been mentioned in this
thread.
Anything else to throw into the ring?
-arturo