Ann W. Harrison wrote:

> >> Excellent analogy. If only the optimizer had enough statistics to tell
> >> hay from needles...
> >
> > Perhaps an idea for future improvement? I guess it wuld require a
> > different kind of index - some kind of hash buckets instead of b-trees?
> >
>
> No, what would be required, I think, is a histogram rather than a
> single number for the selectivity of indexes. Postgres keeps
> ten slot histograms for every field - not just index keys - and
> members of their development group are arguing to increase the
> number of slots to 100.

Interesting. But I suppose it might be of no use in case e.g. every
other value is very common and the rest are very uncommon, and these
happen to be paired together in the slots, e.g. with ten slots:

Value1: 1000 items
Value2: 1 item
Value3: 1000 items
Value4: 1 item
...
Value19: 1000 items
Value20: 1 item

Kind of pathological, but you get the idea. Would this be uncommon
enough for this mechanism to be (very) useful?

Kjell
--
--------------------------------------
Kjell Rilbe
DataDIA AB
E-post: kjell@...
Telefon: 08-761 06 55
Mobil: 0733-44 24 64