> But here is the nasty problem: loads are systematically different.

<snip>

> That's a hard problem.

Query optimization, transaction and concurrency control, recovery in DBMSs
or scheduling algorithms in OSs are/were all hard problems.

Maybe there is someday a solution for a (good) automatic index creation
algorithm based on (longterm) system load. The first idea with a simple
access count statistic will not work. My next step would take user
priorities into account for the calculation of an "attributes include in
index" value.

I had the (weird) idea of concentrating on the design of the database
(relations, primary, unique and foreign key constraints) and letting the
rest be a job for the DBMS.

> I am very sorry about the sweep stuff, incidentally. My fault.
> Next time around, no periodic maintenance.

:-)


Alexander Bauer

Sorry for my bad english. My german is better, for sure.