Re: announce: the "Join Over Histograms" paper

Would I be correct in saying that if you have data that is, or could be, foreign keyed you should not generate histograms, since the CBO default will produce a better map than the formula?

Especially if your foreign keys are to look-up tables.

I found your comments about possible bugs (loop overruns and weighing high values more than low values) very interesting.

--
http://www.freelists.org/webpage/oracle-l