Re: GiST indexing tuples - Mailing list pgsql-performance

From Matthew
Subject Re: GiST indexing tuples
Date
Msg-id Pine.LNX.4.58.0711281607510.3731@aragorn.flymine.org
Whole thread Raw
In response to Re: GiST indexing tuples  (Tom Lane <tgl@sss.pgh.pa.us>)
Responses Re: GiST indexing tuples  ("Matthew T. O'Connor" <matthew@zeut.net>)
List pgsql-performance
On Wed, 28 Nov 2007, Tom Lane wrote:
> Have you looked at contrib/seg/ ?

Yes, I had a pretty good look at that. However, I believe that in order to
use seg's indexes, I would need to put my data into seg's data type, and
reformat my query, as I stated in my original message. What I'm looking
for is a general R-tree (or similar) index that will index multiple
columns of normal data types.

For instance, the normal B-tree index on (a, b) is able to answer queries
like "a = 5 AND b > 1" or "a > 5". An R-tree would be able to index these,
plus queries like "a > 5 AND b < 1".

As far as I can see, it is not possible at the moment to write such an
index system for GiST, which is a shame because the actual R-tree
algorithm is very simple. It's just a matter of communicating both values
from the query to the index code.

Matthew

--
I have an inferiority complex. But it's not a very good one.

pgsql-performance by date:

Previous
From: "Dave Dutcher"
Date:
Subject: Re: Query only slow on first run
Next
From: "Usama Munir Dar"
Date:
Subject: Re: Windows XP selects are very slow