On Tuesday May 23 2006 4:55 pm, Jim C. Nasby wrote:
> Well, I did find one reason not to go ape with this: the
> number of pages analyzed scales with the number of buckets, so
> doubling the statistics target will roughly double the ANALYZE
> time for any table over 6000 pages (though the effect isn't
> linear, see below). There is a small increase in time for a
> small table, but I doubt it's enough for anyone to care:
Are you accounting for your well-primed OS and DB caches? I'd
think a more realistic test would clear those of the target
table between measurements.
Ed