Re: benchmarking the query planner

Simon Riggs <simon@2ndQuadrant.com> writes:
> On Thu, 2008-12-11 at 22:29 +0000, Gregory Stark wrote:
>>> And I would like it even more if the sample size increased according
>>> to table size, since that makes ndistinct values fairly random for
>>> large tables.
>> 
>> Unfortunately _any_ ndistinct estimate based on a sample of the table
>> is going to be pretty random.

> We know that constructed data distributions can destroy the
> effectiveness of the ndistinct estimate and make sample size irrelevant.
> But typical real world data distributions do improve their estimations
> with increased sample size and so it is worthwhile.

This is handwaving unsupported by evidence.  If you've got a specific
proposal what to change the sample size to and some numbers about what
it might gain us or cost us, I'm all ears.
        regards, tom lane