From owner-freebsd-smp Wed Sep 17 12:58:48 1997 Return-Path: Received: (from root@localhost) by hub.freebsd.org (8.8.7/8.8.7) id MAA05990 for smp-outgoing; Wed, 17 Sep 1997 12:58:48 -0700 (PDT) Received: from ganymede.or.intel.com (root@ganymede.or.intel.com [134.134.248.3]) by hub.freebsd.org (8.8.7/8.8.7) with ESMTP id MAA05980 for ; Wed, 17 Sep 1997 12:58:42 -0700 (PDT) Received: from ichips.intel.com (ichips.intel.com [134.134.50.200]) by ganymede.or.intel.com (8.8.6/8.8.5) with ESMTP id NAA05959; Wed, 17 Sep 1997 13:01:10 -0700 (PDT) Received: from pdxlx008.jf.intel.com by ichips.intel.com (8.7.4/jIIIa) id MAA15753; Wed, 17 Sep 1997 12:57:51 -0700 (PDT) Received: from pdxlx008 (localhost [127.0.0.1]) by pdxlx008.jf.intel.com (8.8.7/8.7.3) with ESMTP id MAA05986; Wed, 17 Sep 1997 12:56:41 -0700 (PDT) Message-Id: <199709171956.MAA05986@pdxlx008.jf.intel.com> To: dg@root.com cc: Simon Shapiro , Terry Lambert , missmanp@milo.cfw.com, freebsd-smp@freebsd.org Subject: Re: SMP in FreeBSD 3.x.x In-reply-to: Your message of "Sun, 14 Sep 1997 20:55:03 PDT." <199709150355.UAA24481@implode.root.com> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Date: Wed, 17 Sep 1997 12:56:40 -0700 From: Wayne Scott Sender: owner-freebsd-smp@freebsd.org X-Loop: FreeBSD.org Precedence: bulk -----BEGIN PGP SIGNED MESSAGE----- > This contradicts a paper that was given at a recent Usenix-sponsored > conference that shows that the amount of main memory traffic in P6 systems > is *dramatically* reduced over Pentium systems. I forget the exact ratio, > but it is something like 1/5th. This is entirely due to the superior cache > architecture of the P6. For the huge application/poor hit rate case you have to remember that Pentium does not have a write allocate cache. So a write that misses the cache on a Pentium just writes straight to memory. On a Pentium Pro that line is allocated in the L2 possibly evicting another line to make room. This is more traffic if you will never use the data again. (Rare but it does happen with some applications like TPCC) Dr. Nicely's prime number program (of Pentium FDIV fame) thrashes the Pentium Pro's caches nicely this this regard. Note write allocate is a big win for most programs. Wayne Scott MD6 Architecture wscott@ichips.intel.com Work #: (503) 264-4165 Disclaimer: All views expressed are my own opinions, and not necessarily those of Intel Corporation. -----BEGIN PGP SIGNATURE----- Version: 2.6.2 Comment: Processed by Mailcrypt 3.4, an Emacs/PGP interface iQCVAwUBNCA188R3RVVTfbqpAQFTnAP/fbxPxFUEjdxuNQOExaU5qLJmzuDq76K6 k5MuoEg2iQpoBnjLckPqMTR0Vb5XMwwcjjzng0XpeO9WMtmA2G2CU2ZJAF9WCN5F zYw5DHQGkHY7N+qw26R/ydwSGvsj+3Wqoi9u+mb9AsT3+5L7QpeWUmwrzTZ0vwqw tfLjpozpd0E= =H2KM -----END PGP SIGNATURE-----