To: Simon Riggs <simon@2ndQuadrant.com>
cc: David Kerr <dmk@mr-paradox.net>, pgsql-performance@postgresql.org
Subject: Re: Question on pgbench output 
In-reply-to: <1238914900.5444.292.camel@ebony.2ndQuadrant> 
References: <20090403195302.GA54342@mr-paradox.net>
	<alpine.GSO.2.01.0904031721570.5502@westnet.com>
	<24809.1238799146@sss.pgh.pa.us>
	<20090403233458.GC54342@mr-paradox.net>
	<1238914900.5444.292.camel@ebony.2ndQuadrant>
Comments: In-reply-to Simon Riggs <simon@2ndQuadrant.com>
	message dated "Sun, 05 Apr 2009 08:01:39 +0100"
Date: Sun, 05 Apr 2009 11:46:52 -0400
Message-ID: <4665.1238946412@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Simon Riggs <simon@2ndQuadrant.com> writes:
> On Fri, 2009-04-03 at 16:34 -0700, David Kerr wrote:
>> 400 concurrent users doesn't mean that they're pulling 1.5 megs /
>> second every second.

> There's a world of difference between 400 connected and 400 concurrent
> users. You've been testing 400 concurrent users, yet without measuring
> data transfer. The think time will bring the number of users right down
> again, but you really need to include the much higher than normal data
> transfer into your measurements and pgbench won't help there.

Actually pgbench can simulate think time perfectly well: use its \sleep
command in your script.  I think you can even set it up to randomize the
sleep time.

I agree that it seems David has been measuring a case far beyond what
his real problem is.

			regards, tom lane