Re: improving GROUP BY estimation

On Thu, Mar 3, 2016 at 10:16 PM, Tomas Vondra <tomas(dot)vondra(at)2ndquadrant(dot)com>
wrote:

> So yes, each estimator works great for exactly the opposite cases. But
> notice that typically, the results of the new formula is much higher than
> the old one, sometimes by two orders of magnitude (and it shouldn't be
> difficult to construct examples of much higher differences).
>
> The table also includes the 'average' estimator you propose, but it's
> rather obvious that the result is always much closer to the new value,
> simply because
>
> (small number) + (huge number)
> ------------------------------
> 2
>
> is always much closer to the huge number. We're usually quite happy when
> the estimates are within the same order of magnitude, so whether it's K or
> K/2 makes pretty much no difference.

I believe that Mark means geometrical average, i.e. sqrt((small number) *
(huge number)).

------
Alexander Korotkov
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company