From owner-freebsd-stable@FreeBSD.ORG  Thu Dec 15 20:05:40 2005
Return-Path: <owner-freebsd-stable@FreeBSD.ORG>
X-Original-To: stable@freebsd.org
Delivered-To: freebsd-stable@FreeBSD.ORG
Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125])
	by hub.freebsd.org (Postfix) with ESMTP id A1A2E16A41F
	for <stable@freebsd.org>; Thu, 15 Dec 2005 20:05:40 +0000 (GMT)
	(envelope-from petefrench@ticketswitch.com)
Received: from mail.ticketswitch.com (mail.ticketswitch.com [194.200.93.188])
	by mx1.FreeBSD.org (Postfix) with ESMTP id 3C6D443D64
	for <stable@freebsd.org>; Thu, 15 Dec 2005 20:05:40 +0000 (GMT)
	(envelope-from petefrench@ticketswitch.com)
Received: from [172.16.1.6] (helo=dilbert.firstcallgroup.co.uk)
	by mail.ticketswitch.com with esmtp (Exim 4.52 (FreeBSD))
	id 1EmzMB-0009ah-52; Thu, 15 Dec 2005 20:05:39 +0000
Received: from petefrench by dilbert.firstcallgroup.co.uk with local (Exim
	4.52 (FreeBSD)) id 1EmzMA-0006Rw-U4; Thu, 15 Dec 2005 20:05:38 +0000
To: PeterJeremy@optushome.com.au
In-Reply-To: <20051215182250.GO77268@cirb503493.alcatel.com.au>
Message-Id: <E1EmzMA-0006Rw-U4@dilbert.firstcallgroup.co.uk>
From: Pete French <petefrench@ticketswitch.com>
Date: Thu, 15 Dec 2005 20:05:38 +0000
Cc: stable@freebsd.org
Subject: Re: kernel cpu entries
X-BeenThere: freebsd-stable@freebsd.org
X-Mailman-Version: 2.1.5
Precedence: list
List-Id: Production branch of FreeBSD source code <freebsd-stable.freebsd.org>
List-Unsubscribe: <http://lists.freebsd.org/mailman/listinfo/freebsd-stable>, 
	<mailto:freebsd-stable-request@freebsd.org?subject=unsubscribe>
List-Archive: <http://lists.freebsd.org/pipermail/freebsd-stable>
List-Post: <mailto:freebsd-stable@freebsd.org>
List-Help: <mailto:freebsd-stable-request@freebsd.org?subject=help>
List-Subscribe: <http://lists.freebsd.org/mailman/listinfo/freebsd-stable>,
	<mailto:freebsd-stable-request@freebsd.org?subject=subscribe>
X-List-Received-Date: Thu, 15 Dec 2005 20:05:40 -0000

> I can't see anything in the kernel source code to explain it.  Since
> you don't mention actual times, is the difference statistically
> significant?  (see src/tools/tools/ministat)

Ministat says: Difference at 95.0% confidence

The second set are always smaller than the first set no matter how many
times I run it, so it is repeatable. I only wrote down a few of the raw
results, but here are a set of three outputs from time (real, user, system)
for i686 alone and i586+i686.

	i686:
		496.26	857.54	43.05
		501.00	858.03	42.40
		517.04	857.90	42.91

	i586+i686:
		483.70	852.70	51.77
		484.93	853.54	50.60
		489.26	855.23	46.82

It is a shame I didnt do any without the -j2 on. I suspect that it would show
a slowdown, as the user+system times are always lower on the i686 on its
own. But when running in parallel you actually get a speedup in elapsed
time, even though you are seeing a slowdown on each processor individually.

So does adding in i586 somehow increase the potential for parallelism
somehow ? Thats the only thing I can think of....

-pete.