Message-ID: <4C0B52C3.4050901@postnewspapers.com.au>
Date: Sun, 06 Jun 2010 15:48:19 +0800
From: Craig Ringer <craig@postnewspapers.com.au>
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US;
	rv:1.9.1.9) Gecko/20100423 Lightning/1.0b1 Thunderbird/3.0.4
MIME-Version: 1.0
To: pgsql-performance@postgresql.org
Subject: Re: How filesystems matter with PostgreSQL
References: <4C08EEDF.7090403@mtu.net> <4C0AD180.90607@2ndquadrant.com>
	<4C0AD280.5020105@mtu.net> <4C0AD547.2040308@2ndquadrant.com>
	<4C0AD7DF.1030705@mtu.net>
	<4C0B455C.8050902@cheapcomplexdevices.com>
In-Reply-To: <4C0B455C.8050902@cheapcomplexdevices.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit

On 06/06/10 14:51, Ron Mayer wrote:
> Jon Schewe wrote:
>
>> OK, so if I want the 15 minute speed, I need to give up safety (OK in
>> this case as this is just research testing), or see if I can tune
>> postgres better.
>
> Depending on your app, one more possibility would be to see if you
> can re-factor the application so it can do multiple writes in parallel
> rather than waiting for each one to complete.   If I understand right,
> then many transactions could potentially be handled by a single fsync.

By using a commit delay, yes. (see postgresql.conf). You do open up the 
risk of losing transactions committed within the commit delay period, 
but you don't risk corruption like you do with fsync.

Sometimes you can also batch work into bigger transactions. The classic 
example here is the usual long stream of individual auto-committed 
INSERTs, which when wrapped in an explicit transaction can be vastly 
quicker.

--
Craig Ringer