To: mark@mark.mielke.cc
Cc: markus@m-bass.com, pgsql-performance@postgresql.org
Subject: Re: When to do a vacuum for highly active table 
In-reply-to: <20050830220503.GA27740@mark.mielke.cc> 
References: <20050830211356.48138.qmail@web31606.mail.mud.yahoo.com>
	<24875.1125437357@sss.pgh.pa.us>
	<20050830220503.GA27740@mark.mielke.cc>
Comments: In-reply-to mark@mark.mielke.cc
	message dated "Tue, 30 Aug 2005 18:05:03 -0400"
Date: Tue, 30 Aug 2005 19:05:04 -0400
Message-ID: <25546.1125443104@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

mark@mark.mielke.cc writes:
> I think he means splitting it vertically, instead of horizontally, and
> it sounds like an excellent idea, if a large enough portion of each
> record is in fact mostly fixed. Otherwise, PostgreSQL is copying data
> multiple times, only to have the data expire as part of a dead row.

Only up to a point.  Fields that are wide enough to get toasted
out-of-line (multiple Kb) do not get physically copied if there's
a row update that doesn't affect them.  We don't really have enough
information about his table to guess whether there's any point in
manually partitioning the columns, but my leaning would be "probably
not" --- the overhead in joining the resulting two tables would be
high enough that you'd need a heck of a big improvement to justify it.

			regards, tom lane