Message-ID: <52F93098.2090002@vmware.com>
Date: Mon, 10 Feb 2014 22:03:36 +0200
From: Heikki Linnakangas <hlinnakangas@vmware.com>
Organization: VMware
User-Agent: Mozilla/5.0 (X11; Linux x86_64;
 rv:17.0) Gecko/20131103 Icedove/17.0.10
MIME-Version: 1.0
To: M Putz <puma@c007.de>
CC: pgsql-performance@postgresql.org
Subject: Re: Strange performance boost with random()
References: <E1WCttM-0006Np-Nj@malur.postgresql.org>
 <alpine.DEB.2.02.1402102007580.14849@skaj>
In-Reply-To: <alpine.DEB.2.02.1402102007580.14849@skaj>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk
Sender: pgsql-performance-owner@postgresql.org

On 02/10/2014 09:52 PM, M Putz wrote:
>
> Hello,
>
> While analyzing performance, we encountered the following phenomenon,
>
>     SELECT sum(pow(.5*generate_series,.5))
>     FROM generate_series(1,1000000);
>
> is much much (a hundred times) slower than
>
>     SELECT sum(pow(random()*generate_series,.5))
>     FROM generate_series(1,1000000);
>
> and asymptotic difference is even more astounding.
> This seems counter-intuitive, considering the cost of
> an additional random() call instead of a constant factor.
> What are the reasons for this strange performance boost?

Different data type. The first uses numeric, which is pretty slow for 
doing calculations. random() returns a double, which makes the pow and 
sum to also use double, which is a lot faster.

To see the effect, try these variants:

SELECT sum(pow(.5::float8 * generate_series,.5))
FROM generate_series(1,1000000);

SELECT sum(pow(random()::numeric * generate_series,.5))
FROM generate_series(1,1000000);

- Heikki


-- 
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance