Message-ID: <4EC926C9.1030409@wizmail.org>
Date: Sun, 20 Nov 2011 16:11:53 +0000
From: Jeremy Harris <jgh@wizmail.org>
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US;
	rv:1.8.1.21) Gecko/20090320 Fedora/2.0.0.21-1.fc9
	Thunderbird/2.0.0.21 Mnenhy/0.7.5.0
MIME-Version: 1.0
To: pgsql-performance@postgresql.org
Subject: Re: external sort performance
References: 
 <CAKuK5J3taH85P6zc1j63mkwiNXH2nfJ5ER6R1+Vk5iG_Cq-cFw@mail.gmail.com>
	<4EC906FE.5040902@wizmail.org>
	<CAKuK5J2Z5d=ADthwGvaxu2OVJUHMwsHQSJvpmPtskRsZAkHKzQ@mail.gmail.com>
In-Reply-To: 
 <CAKuK5J2Z5d=ADthwGvaxu2OVJUHMwsHQSJvpmPtskRsZAkHKzQ@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit

On 2011-11-20 15:00, Jon Nelson wrote:
>  Do you happen to recall if disk I/O is
> counted as user or system time?

Neither, traditionally.  Those times are cpu times;
they only account for what the cpu was doing.
The disks could be working in parallel as a result
of cpu actions, and probably were - but the software
found work to do for the cpu.   You'd want to be
looking at iostat during the run to see how busy the
disks were.

As to why it takes 16 hours cpu to do the external
version but only 34 minutes for internal - some of that
will be down to data-shuffling in- and out- of disk files
which is nonetheless accounted to user-mode cpu time,
but some will be the plain inefficiency of the external
version having to effectively do work over many times
because it can't have a complete view of the problem at
hand at any one time.

-- 
Jeremy