From owner-freebsd-smp  Fri May  2 09:06:55 1997
Return-Path: <owner-smp>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.5/8.8.5) id JAA07996
          for smp-outgoing; Fri, 2 May 1997 09:06:55 -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 JAA07979;
          Fri, 2 May 1997 09:06:16 -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 KAA14190;
	Fri, 2 May 1997 10:05:33 -0600 (MDT)
Message-Id: <199705021605.KAA14190@Ilsa.StevesCafe.com>
X-Mailer: exmh version 2.0gamma 1/27/96
From: Steve Passe <smp@csn.net>
To: "Jonathan M. Bresler" <jmb@FreeBSD.ORG>
cc: andreas@klemm.gtn.com (Andreas Klemm), chuckr@mat.net, WH@ODS.de,
        smp@FreeBSD.ORG, everybodyunix@wup.de, jk@ct.heise.de
Subject: Re: Dual PPro Mainboard f. SCO SMP 
In-reply-to: Your message of "Fri, 02 May 1997 07:24:30 PDT."
             <199705021424.HAA02783@hub.freebsd.org> 
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Date: Fri, 02 May 1997 10:05:33 -0600
Sender: owner-smp@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

Hi,

> > 	- how much performance (loss) you get, when 2 CPU's have
> > 	  to share the same 2nd Level cache (Pentium/MMX case)
> 
> this one is of particular interest to me, anyone set up to test this?
> given numbers for a Pentium dual proc system and a PetniumPro dual proc
> i can normalize the output ;)

both with COPTFLAGS = -O


dual P5-133mHz with shared 512kb external cache.
wide 7200rpm SCSI disk.
 # make clean && make depend && time make -j8
 318.39s real  458.19s user  126.32s system 

dual P6-200mHz each with 512kb internal cache (P6-166 overclocked to 200mHz):
older narrow 5400 SCSI disk.
/usr/src mounted async,noatime.
 # make clean && make depend && time make -j8
 104.08s real  129.70s user   53.88s system 

---
on a similar note, I played with the share/mk/* files trying to get
"-j8" into them for a world build.  I kept hitting places where the
Makefile/*.mk files didn't define dependancies closely enough to avoid
build errors (links made b4 files copied, etc.).  But by timing
the build of lower level sub-dirs I saw 25%+ speedup where it worked.
This was just by adding "-j8" to bsd.subdir.mk.  If someone who had
some experience with these guys were to work a little at it I think
it could be made to fly.  -j4 speeds up even UP kernel builds, so it
would be useful for both UP and SMP worlds.  It would be nice if it
could be done so that it was an /etc/make.conf setting.

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