Re: GiST opclass and varlena

Hi all,

The first version of prefix code is now released at pgfoundry, last bug
squashed by Jordan, who did his own prefix gist opclass implementation before
I did, but didn't publish at this time. The plan is to integrate is work into
the prefix solution, as his lookups are faster than mine, and the code base
simpler but limited to varchar(15).

You can get the prefix_range datatype and prefix search GiST opclass here:
http://cvs.pgfoundry.org/cgi-bin/cvsweb.cgi/prefix/prefix/
http://prefix.projects.postgresql.org/README.html
http://pgfoundry.org/frs/?group_id=1000352

Le mardi 25 mars 2008, Dragan Zubac a écrit :
> Some poorly measurement showed some 60-80 matching/sec with this
> algorithm of matching prefices and a couple of concurrent database
> connections.

On a workstation class hardware with a 11966 rows prefix table, 25000 random
10 digit numbers where matched in less than 5s, having an average lookup time
of about 0.2ms, or 5000 matches per second.
http://prefix.projects.postgresql.org/TESTS.html

If you need some speedups on prefix matching but are not willing to install
any C-code additional module, please check this blog entry:
http://www.depesz.com/index.php/2008/03/04/searching-for-longest-prefix/

The cvs and tarball version of prefix have been tested on PostgreSQL versions
8.2 and 8.3, figures from tests were obtained with 8.3, and debian packages
are available for 8.3 only at the moment.

Regards,
--
dim