Re: Statistics and selectivity estimation for ranges

On 04.08.2012 12:31, Alexander Korotkov wrote:
> Hackers,
>
> attached patch is for collecting statistics and selectivity estimation for
> ranges.
>
> In order to make our estimations accurate for every distribution of
> ranges, we would collect 2d-distribution of lower and upper bounds of range
> into some kind of 2d-histogram. However, this patch use some simplification
> and assume distribution of lower bound and distribution of length to be
> independent.

Sounds reasonable. Another possibility would be to calculate the average 
length for each lower-bound bin. So you would e.g know the average 
length of values with lower bound between 1-10, and the average length 
of values with lower bound between 10-20, and so forth. Within a bin, 
you would have to assume that the distribution of the lengths is fixed.

PS. get_position() should guard against division by zero, when subdiff 
returns zero.

-- 
   Heikki Linnakangas
   EnterpriseDB   http://www.enterprisedb.com