To: sid@etsy.com
cc: pgsql-performance@postgresql.org
Subject: Re: How can fixed and variable width columns perform similarly? 
In-reply-to: <50516.209.51.183.100.1177690744.squirrel@mailbox.etsy.com> 
References: <50444.209.51.183.100.1177687911.squirrel@mailbox.etsy.com>
	<15873.1177691157@sss.pgh.pa.us>
	<50516.209.51.183.100.1177690744.squirrel@mailbox.etsy.com>
Comments: In-reply-to "Siddharth Anand" <sid@etsy.com>
	message dated "Fri, 27 Apr 2007 09:19:04 -0700"
Date: Fri, 27 Apr 2007 13:12:47 -0400
Message-ID: <16453.1177693967@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

"Siddharth Anand" <sid@etsy.com> writes:
> My question wasn't phrased clearly. Oracle exhibits a performance
> degradation for very large-sized fields (CLOB types that I equate to
> PostGres' text type) when compared with the performance of field types
> like varchar that handle a max character limit of a few thousand bytes in
> Oracle.

> It sounds like PostGres doesn't exhibit this same difference. I wanted to
> understand how this could be and whether there was a trade-off.

Ah.  Well, the answer is that we change behavior dynamically depending
on the size of the particular field value, instead of hard-wiring it to
the declared column type.  It sounds like Oracle's CLOB might be doing
about the same thing as an out-of-line "toasted" field value in
Postgres.  In PG, text and varchar behave identically except that
varchar(N) adds an insert-time check on the length of the field value
--- but this is just a constraint check and doesn't have any direct
influence on how the value is stored.

			regards, tom lane