At 04:25 PM 7/19/00 +1000, Ungod wrote:

> I'd like to
>propose for discussion an addition to the engine to over time gather a
>history of the most used order by's and the most used where criteria to
>create either an "automatic indexing" feature or even just a "what are the
>best indexes i should use on this dataset". over time, being history based
>of course, it would refine itself the more the database gets used.
>Thoughts? Comments?
>

It's an idea that a take out of the closet every decade or so, ponder,
and put back in the closet. A schema to log record selection expressions
and related cardinalities for offline analysis is certainly feasible.

The standard answer, however, is the following scenario: Monday through
Friday the database observers itself pumping transactions and decides
to reorganize itself Friday night. Saturday the week end batch streams
start, optimized for pumping transactions and pessimized for batch.
Sunday night, the database realizes it has made a dreadful mistakes
and re-organizes itself optimized for batch and pessimized for
transactions. Sigh. Can't win. Database gives up the automatic
optimization, marries a web server, and retires to the Bahama to drive
illegal web gambling.

Jim Starkey