Re: rtree/gist index taking enormous amount of space in 8.2.3

The application need is to determine genomic features present in a
user-defined portion of a chromosome. My guess is that features (boxes)
are overlapping along a line (chromosome), and there is a need to
represent them as being stacked. Since I'm not certain of its exact
use, I've emailed the application owner to find the motivation as to why
a geometric index structure is used, and why the boxes are tall and
overlapping. As a side note, the data model for our application is
based on a popular bioinformatics open source project called chado.

Thanks,
Tom
-----Original Message-----
From: Tom Lane [mailto:tgl(at)sss(dot)pgh(dot)pa(dot)us]
Sent: Friday, June 29, 2007 2:38 PM
To: Dolafi, Tom
Cc: pgsql-performance(at)postgresql(dot)org; Oleg Bartunov; Teodor Sigaev
Subject: Re: [PERFORM] rtree/gist index taking enormous amount of space
in 8.2.3

"Dolafi, Tom" <dolafit(at)janelia(dot)hhmi(dot)org> writes:
> In the mean time I've dropped the index which has resulted in overall
> performance gain on queries against the table, but we have not tested
> the part of the application which would utilize this index.

I noted that with the same (guessed-at) distribution of fmin/fmax, the
index size remains reasonable if you change the derived boxes to

CREATE OR REPLACE FUNCTION boxrange(integer, integer)
RETURNS box AS
'SELECT box (point($1, $1), point($2, $2))'
LANGUAGE 'sql' STRICT IMMUTABLE;

which makes sense from the point of view of geometric intuition: instead
of a bunch of very tall, mostly very narrow, mostly overlapping boxes,
you have a bunch of small square boxes spread out along a line. So it
stands to reason that a geometrically-motivated index structure would
work a lot better on the latter. I don't know though whether your
queries can be adapted to work with this. What was the index being used
for, exactly?

regards, tom lane