"Ann W. Harrison" <aharrison@...> wrote:

>
> >
> > I'm pretty sure PG used a hash aggregate which really performs better

than a

> > sort aggregate.
>
> Ok, thanks, I thought there was something like that.

Well, it doesn't make big difference. I've performed a quick test with FB
1.5.2 and PG 8.0 on win32. On second run FB is faster than PG, except the
GROUP BY clause (where PG slightly wins). But the first run (when our page
cache is empty) makes a very big difference: PG is about 20x faster than FB.
I don't believe in miracles, so I suspect some in advance database caching
on the PG side, but neither of the PG backend processes occupied more than
10MB of RAM whilst the test table was 50M. I think this should be
investigated a bit further.


Dmitry