DomainKey-Signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma;
	h=message-id:date:from:user-agent:mime-version:to:cc:subject
	:references:in-reply-to:content-type:content-transfer-encoding;
	b=CrMFMhiEDbP+mdgViXF+0gmRnFsyl9Kkqyspao64nGXk5oU5oIyi/+fZpcI3t0q3iJ
	JOEbg9FnkOiXLrvVqLBOTYHPztN4B6vO1LqudvEPgL7dI+xnjhk0OZcG9PyVOCjN8GYD
	LlZ+dFTuxGnUMyBVUV4vrNfbFQlhz65f0mUAU=
Message-ID: <4C2492F9.2020706@gmail.com>
Date: Fri, 25 Jun 2010 13:28:57 +0200
From: Yeb Havinga <yebhavinga@gmail.com>
User-Agent: Thunderbird 2.0.0.24 (Windows/20100228)
MIME-Version: 1.0
To: Rajesh Kumar Mallah <mallah.rajesh@gmail.com>
CC: Kevin Grittner <Kevin.Grittner@wicourts.gov>,
	pgsql-performance@postgresql.org
Subject: Re: sudden spurt in swap utilization (was:cpu bound postgresql
 setup.)
References: <AANLkTinDBPQL-SjLaCr7AVgzouWbRrjwA6MAgBih_ciy@mail.gmail.com>
In-Reply-To: <AANLkTinDBPQL-SjLaCr7AVgzouWbRrjwA6MAgBih_ciy@mail.gmail.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Rajesh Kumar Mallah wrote:
> A scary phenomenon is being exhibited by the server , which is the server
> is slurping all the swap suddenly
>   
>  8  1 4192912 906164   6100 27873640    0    0  2277   858 13440 16235
> 63  8 19 10  0
>
> I reduced work_mem from 4GB to 2GB to 512MB (now). I clearly remember that this
> abnormal consumption of swap was NOT there even when work_mem was 4GB.
> eg during happier times swap utilisation was: http://pastebin.com/bnE1pFZ9
>   
> the question is whats making postgres slurp the swap? i am posting my
> current postgresql.conf
> once again.
>
> # cat postgresql.conf  | grep -v "^\s*#" | grep -v "^\s*$"
> listen_addresses = '*'          # what IP address(es) to listen on;
> port = 5432                             # (change requires restart)
> max_connections = 300                   # (change requires restart)
>   
Hello Rajesh,

In constrast with e.g. shared_buffers and effective_cache_size, work_mem 
is amount of memory per 'thing' (e.g. order/group by) that wants some 
working memory, so even a single backend can use several pieces of 
work_mem memory.

Looking at your postgresql.conf, other memory values seem a bit too high 
as well for a 32GB ram server. It is probably a good idea to use pgtune 
(on pgfoundry) to get some reasonable ball park settings for your hardware.

regards,
Yeb Havinga