DomainKey-Signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma;
	h=message-id:date:from:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references;
	b=PB0M05NNwGWl2s6ii6wYhAQWLERjdmd+NHMXSMtLoe3RiIUULQORdbtIJ7qwZPFtdI/abMebex7ZZS+2oB39GHrFps+CBbQgvhZEndCgAy9/JuKCK1LDn6+z0Xls206KSKhyzuo42D8RLDTTnrB2goUnhY/CjLtgTPUVtaowSO0=
Message-ID: <1d4e0c10805220638x64f80609ob9a6fe00e4587362@mail.gmail.com>
Date: Thu, 22 May 2008 15:38:25 +0200
From: "Guillaume Smet" <guillaume.smet@gmail.com>
To: "Tom Lane" <tgl@sss.pgh.pa.us>
Subject: Re: Index creation time and distribution
Cc: pgsql-performance <pgsql-performance@postgresql.org>
In-Reply-To: <29529.1211462076@sss.pgh.pa.us>
MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
References: <1d4e0c10805220532v514c2bd5je4c6c20ceb485495@mail.gmail.com>
	<29529.1211462076@sss.pgh.pa.us>

On Thu, May 22, 2008 at 3:14 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Do you have maintenance_work_mem set large enough that the index
> creation sort is done in-memory?  8.1 depends on the platform's qsort
> and a lot of them are kinda pessimal for input like this.

FWIW, it's a 32 bits CentOS 4.6 box.

maintenance_work_mem is set to 256 MB and the size of the index is 400 MB.

Should I try to raise it up to 512 MB? The server only has 2GB of RAM
so it seems a bit high.

> 8.2 (which uses our own qsort) seems to perform better in a quick
> test.

Mmmmh OK. I was considering an upgrade to 8.3 in the next months anyway.

Do we agree that in the case of unnamed prepared statement, 8.3 plans
the query after the BIND? The partial index seems to be a better
solution anyway, considering that it's 12 MB vs 400 MB.

Thanks.

-- 
Guillaume