From: Chris Browne <cbbrowne@acm.org>
Subject: Re: performance of like queries
Date: Tue, 02 Oct 2007 17:55:08 -0400
Organization: cbbrowne Computing Inc
Lines: 31
Message-ID: <60abr1mbb7.fsf@dba2.int.libertyrms.com>
References: <200710021131.46010.kevin@kevinkempterllc.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
User-Agent: Gnus/5.1008 (Gnus v5.10.8) XEmacs/21.4.20 (linux)
Cancel-Lock: sha1:FkEalzHaAiIwafqlMueCgqZ2qZI=
To: pgsql-performance@postgresql.org

kevin@kevinkempterllc.com (Kevin Kempter) writes:
> any suggestions for improving "LIKE '%text%'" queries?

If you know that the 'text' portion of that query won't change, then
you might create a partial index on the boolean condition.

That is, 

  create index index_foo_text on my_table (tfield) where (tfield like '%text%');

I somehow doubt that is the case; more likely you want to be able to
search for:
   select * from my_table where tfield like '%this%';
   select * from my_table where tfield like '%that%';
   select * from my_table where tfield like '%the other thing%';

There are basically three choices, at that point:

1.  Get more memory, and hope that you can have all the data get
cached in memory.

2.  Get more better disk, so that you can scan the table faster on
disk.

3.  Look into tsearch2, which provides a full text search capability.
-- 
(format nil "~S@~S" "cbbrowne" "linuxdatabases.info")
http://cbbrowne.com/info/x.html
"We're  born with a  number  of  powerful instincts,  which are  found
across all cultures.   Chief amongst these are  a dislike of snakes, a
fear of falling, and a hatred of popup windows" -- Vlatko Juric-Kokic