> Ok, I can reproduce the issue with that. The index is only 4MB in size
> when I populate it with random data (vs. 15 MB with your data). The
> command I used is:
>
> INSERT INTO cubtest SELECT cube(random(), random()) FROM
> generate_series(1,20000);
>
> My guess is that the picksplit algorithm performs poorly with that data.
> Unfortunately, I have no idea how to improve that.
One of idea is add sorting of Datums to be splitted by cost of insertion. It's
implemented in intarray/tsearch GiST indexes.
Although I'm not sure that it will help but our researches on Guttman's
picksplit algorimth show significant improvements.
--
Teodor Sigaev E-mail: teodor@sigaev.ru
WWW: http://www.sigaev.ru/