From owner-freebsd-questions Thu Apr 6 23:55:18 2000 Delivered-To: freebsd-questions@freebsd.org Received: from isy.liu.se (isy.liu.se [130.236.48.10]) by hub.freebsd.org (Postfix) with ESMTP id 01DE537C237 for ; Thu, 6 Apr 2000 23:55:06 -0700 (PDT) (envelope-from mj@isy.liu.se) Received: from lagrange.isy.liu.se (lagrange.isy.liu.se [130.236.49.127]) by isy.liu.se (8.10.0/8.10.0) with ESMTP id e376sYm17661; Fri, 7 Apr 2000 08:54:35 +0200 (MET DST) Message-ID: X-Mailer: XFMail 1.3 [p0] on FreeBSD X-Priority: 3 (Normal) Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: 8bit MIME-Version: 1.0 In-Reply-To: Date: Fri, 07 Apr 2000 08:54:34 +0200 (CEST) From: Micke Josefsson To: R Joseph Wright Subject: Re: smp motherboards/concepts Cc: freebsd-questions@FreeBSD.ORG, Alfred Perlstein Sender: owner-freebsd-questions@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG On 06-Apr-00 R Joseph Wright wrote: > On Thu, 6 Apr 2000, Alfred Perlstein wrote: > >> * R Joseph Wright [000406 12:07] wrote: >> > I know FreeBSD supports multiple processors, but how many? What >> > motherboard(s) support more than two, for example using a pentium pro? >> >> People have been known to use FreeBSD in quad processor systems. >> >> > >> > Is the concept of SMP similar to beowulf clusters? I've read a little >> > about those, and as I understand, simply clustering many computers >> > together >> > does not guarantee a faster system, if the software is not especially >> > compiled to handle it. Is this also true of SMP? I'm specifically >> > thinking of using an SMP system for doing raytracing. >> >> The SMP you get with FreeBSD would be multiple processors in a single >> system, unless you can break down your raytracing application into >> multiple processes to act on different threads at the same time you >> won't see any improvement. > > Is it only useful for running many applications at the same time? You cant get one process to use both processors at the same time (unless you actually program it that way). But if you run povray for example, you can split a picture up in two parts and start two copies of povray and the job will be finsihed in half the time. Povray scales well with a beowulf cluster too: At usenix last year a five machine beowulf was shown to do just this, in this case the rendering time was cut down to about one fifth. With beowulf systems, performance is best if the amount of data between processes in separate computers can be kept to a minimum. The overhead of yanking data over TCP/IP is much, much, much slower than reading your own machines memory, as you can do with threads. But that of course, requires special tinkering with the source code. You may want to look at MPI in the ports if you're into hacking your own code. ---------------------------------- Michael Josefsson, MSEE mj@isy.liu.se This message was sent by XFMail running on FreeBSD 3.4 ---------------------------------- To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message