Message-ID: <48DD6BAB.8040302@mib-infotech.co.nz>
Date: Sat, 27 Sep 2008 11:09:31 +1200
From: John Huttley <John@mib-infotech.co.nz>
User-Agent: Thunderbird 2.0.0.16 (Windows/20080708)
MIME-Version: 1.0
To: Greg Smith <gsmith@gregsmith.com>
CC: pgsql-performance@postgresql.org
Subject: Re: Slow updates, poor IO
References: <48DBE587.3050406@mib-infotech.co.nz>
	<Pine.GSO.4.64.0809251750130.16879@westnet.com>
In-Reply-To: <Pine.GSO.4.64.0809251750130.16879@westnet.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Hi Greg,

I've got 32M shared on a 1G machine and 16 checkpoint segments.
I'll run some tests against 64 segments and see what happens.

Your previous postings were extremely helpful wrt the MVCC issue.
I thank you!

-john


Greg Smith wrote:
> On Fri, 26 Sep 2008, John Huttley wrote:
>
>> running    update file set perms='0664' took about 10 mins
>
> What do you have checkpoint_segments and shared_buffers set to?  If 
> you want something that's doing lots of updates to perform well, you 
> need to let PostgreSQL have a decent size chunk of memory to buffer 
> the index writes with, so it's more likely they'll get combined into 
> larger and therefore more easily sorted blocks rather than as more 
> random ones.  The randomness of the writes is why your write rate is 
> so slow.  You also need to cut down on the frequency of checkpoints 
> which are very costly on this type of statement.
>
> Also:  which version of PostgreSQL?  8.3 includes an improvement aimed 
> at updates like this you might benefit from.
>
> -- 
> * Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD
>