From owner-freebsd-hardware  Fri May 30 18:39:18 1997
Return-Path: <owner-hardware>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.5/8.8.5) id SAA20873
          for hardware-outgoing; Fri, 30 May 1997 18:39:18 -0700 (PDT)
Received: from Ilsa.StevesCafe.com (Ilsa.StevesCafe.com [205.168.119.129])
          by hub.freebsd.org (8.8.5/8.8.5) with ESMTP id SAA20860;
          Fri, 30 May 1997 18:39:03 -0700 (PDT)
Received: from Ilsa.StevesCafe.com (localhost [127.0.0.1])
	by Ilsa.StevesCafe.com (8.8.5/8.8.5) with ESMTP id TAA24629;
	Fri, 30 May 1997 19:38:57 -0600 (MDT)
Message-Id: <199705310138.TAA24629@Ilsa.StevesCafe.com>
X-Mailer: exmh version 2.0gamma 1/27/96
From: Steve Passe <smp@csn.net>
To: "Jin Guojun[ITG]" <jin@george.lbl.gov>
cc: hardware@FreeBSD.ORG, smp@FreeBSD.ORG
Subject: Re: SMP (3.0-970527-SNAP) on dual pentium machines 
In-reply-to: Your message of "Fri, 30 May 1997 17:27:33 PDT."
             <199705310027.RAA29428@george.lbl.gov> 
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Date: Fri, 30 May 1997 19:38:57 -0600
Sender: owner-hardware@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

Hi,

> I have two different dual pentium motherboards:
> 	1)	ASUS P/I-P65UP5 with C-P55T2D
> 	2)	TYAN S1563D
> 
> neither of them really gets worked for SMP. Here is the testing result by
> using pthread library -lc_r:
> 
> -----------------------
> 
> FreeBSD with single CPU:
> 60 /data/src: vfft mri.b128 > /dev/null
> 9.7u 0.1s 0:09.95 99.3% 97+11132k 0+0io 0pf+0w
> 
> FreeBSD with dual CPU + pthread (failure SMP):
> 62 /data/src: vfft.mt < mri.b128 > /dev/null
> 9.5u 0.2s 0:09.83 99.1% 77+11822k 0+0io 0pf+0w
> 
> 
> -----------------------
> 
> Solaris 2.5.1.u8 with single CPU:
> 67 /home/data/src: vfft mri.b128 > /dev/null
> 12.0u 0.0s 0:13 87% 0+0k 0+0io 0pf+0w
> 
> Solaris 2.5.1.u8 with dual CPU + thread (SMP):
> 69 /home/data/src: vfft.mt mri.b128 > /dev/null
> 12.0u 0.0s 0:07 168% 0+0k 0+0io 0pf+0w
> 
> -----------------------
> 
> We can see that Solaris does make SMP work, but FreeBSD 3.0-SNAP does not.
> Does this fail at pthread library (-lc_r)?  I have not use fork() for a while.
> I will test it sometimes later.
> 
> I saw some message early that said SMP works. Would some one please tell me
> how was the SMP tested (fork() or pthread)?

what you are demonstrating is the fact that we do not yet have kernel threads
working.  The threads provided by our libc_r all run in one process, thus no 
improvement in thruput for your test.

Note that FreeBSD does much better (if I'm reading thse right) than Solaris
in the single CPU test.  Once we have kernel threads we should also do 
substantially better than Solaris in the MP test.

FreeBSD is using all CPUs, and any situation that benefits from running
multiple processes at the same time should see real differences.  Check out:

http://www.freebsd.org/~fsmp/SMP/akgraph-a/graph1.htm

Note that I can now build an SMP kernel in 95 seconds (no X running) on
a dual P6-200x512.

--
Steve Passe	| powered by 
smp@csn.net	|            Symmetric MultiProcessor FreeBSD