Message-ID: <4DDAB2CD.2050104@fuzzy.cz>
Date: Mon, 23 May 2011 21:17:33 +0200
From: Tomas Vondra <tv@fuzzy.cz>
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US;
	rv:1.9.2.13) Gecko/20101219 Lightning/1.0b3pre Thunderbird/3.1.7
MIME-Version: 1.0
To: pgsql-performance@postgresql.org
Subject: Re: Performance degradation of inserts when database size
 grows
References: <BANLkTi=VKBmRLVLDjy8qxpWx_6-rmbUaXg@mail.gmail.com>
	<4DDA617A.6090105@peak6.com>	<4DDAAB8F.5030508@fuzzy.cz>
	<BANLkTikEGR0Mz-bRbk4eXrFtgLt6vC4t_w@mail.gmail.com>
In-Reply-To: <BANLkTikEGR0Mz-bRbk4eXrFtgLt6vC4t_w@mail.gmail.com>
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit

Dne 23.5.2011 21:05, Robert Haas napsal(a):
> On Mon, May 23, 2011 at 2:46 PM, Tomas Vondra <tv@fuzzy.cz> wrote:
>> Really? He already has 64 checkpoint segments, which is about 1GB of
>> xlog data. The real problem is that the amount of buffers to write is
>> constantly growing. At the beginning there's 62861 buffers (500MB) and
>> at the end there's 137657 buffers (1GB).
>>
>> IMHO increasing the number of checkpoint segments would make this
>> disruption even worse.
> 
> Maybe - but it would also make the checkpoints less frequent, which
> might be a good thing.
> 
>> In 9.1 there's a feature that spreads checkpoint writes, but with 8.4
>> that's not possible.
> 
> What feature are you referring to here?  Checkpoint spreading was
> added in 8.3, IIRC.

You're absolutely right, I was talking about

  checkpoint_completion_target

and it was added in 8.3. Your memory is obviously better than mine.

Tomas