Message-ID: <4BB0DB0D.80208@2ndquadrant.com>
Date: Mon, 29 Mar 2010 12:53:33 -0400
From: Greg Smith <greg@2ndquadrant.com>
User-Agent: Thunderbird 2.0.0.24 (X11/20100317)
MIME-Version: 1.0
To: "Campbell, Lance" <lance@illinois.edu>
CC: pgsql-performance@postgresql.org
Subject: Re: How much memory is PostgreSQL using
References: 
 <B10E6810AC2A2F4EA7550D072CDE876004B2E218@SAB-FENWICK.sab.uiuc.edu>
In-Reply-To: 
 <B10E6810AC2A2F4EA7550D072CDE876004B2E218@SAB-FENWICK.sab.uiuc.edu>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Campbell, Lance wrote:
>
> Or is there some way to ask PostgreSQL how much memory are you using 
> to cache disk blocks currently?
>

You can install contrib/pg_buffercache into each database and count how 
many used blocks are there.  Note that running queries using that 
diagnostic tool is really intensive due to the locks it takes, so be 
careful not to do that often on a production system.


> When you do a PG_DUMP does PostgreSQL put the disk blocks into shared 
> buffers as it runs?
>

To some extent.  Most pg_dump activity involves sequential scans that 
are reading an entire table.  Those are no different from any other 
process that will put disk blocks into shared_buffers.  However, that 
usage pattern makes pg_dump particularly likely to run into an 
optimization in 8.3 and later that limits how much of shared_buffers is 
used when sequentially scanning a large table.  See P10 of 
http://www.westnet.com/~gsmith/content/postgresql/InsideBufferCache.pdf 
for the exact implementation.  Basically, anything bigger than 
shared_buffers / 4 uses a 256K ring to limit its cache use, but it's a 
little more complicated than that.

-- 
Greg Smith  2ndQuadrant US  Baltimore, MD
PostgreSQL Training, Services and Support
greg@2ndQuadrant.com   www.2ndQuadrant.us