Re: What about utility to calculate planner cost constants? - Mailing list pgsql-performance

On Mon, 21 Mar 2005 14:59:56 -0800, Josh Berkus <josh@agliodbs.com> wrote:
> > If by not practical you mean, "no one has implemented a multivariable
> > testing approach," I'll agree with you. But multivariable testing is
> > definitely a valid statistical approach to solving just such problems.
> Well, not practical as in:  "would take either $10 million in equipment or
> 10,000 hours or both"

I think you don't need EXPLAIN ANALYZE each query with different GUCs,
you would only need EXPLAIN most of the times (which is way quicker).
Once you get 'near' actual values, you would do EXECUTE ANALYZE to
verify the variables.

   Regards,
      Dawid