"Jim C. Nasby" <jnasby@xxxxxxxxxxxxx> writes: > On Mon, May 22, 2006 at 06:43:22PM -0400, Tom Lane wrote: >> It would be nice to have some *evidence*, not unsupported handwaving. > If someone has an idea on how to actually get that evidence, I'm all > ears. Well, for example, actually measuring the planner overhead from larger pg_statistic entries would be interesting. Plus how much more time ANALYZE takes to generate the entries. (I'm afraid that ANALYZE is probably worse-than-linear CPU-wise, but it may be dominated by disk access.) regards, tom lane