From owner-freebsd-hackers Mon Oct 6 08:27:48 1997 Return-Path: Received: (from root@localhost) by hub.freebsd.org (8.8.7/8.8.7) id IAA04751 for hackers-outgoing; Mon, 6 Oct 1997 08:27:48 -0700 (PDT) (envelope-from owner-freebsd-hackers) Received: from terra.Sarnoff.COM (terra.sarnoff.com [130.33.11.203]) by hub.freebsd.org (8.8.7/8.8.7) with SMTP id IAA04740 for ; Mon, 6 Oct 1997 08:27:44 -0700 (PDT) (envelope-from rminnich@Sarnoff.COM) Received: (from rminnich@localhost) by terra.Sarnoff.COM (8.6.12/8.6.12) id LAA13654; Mon, 6 Oct 1997 11:27:01 -0400 Date: Mon, 6 Oct 1997 11:27:00 -0400 (EDT) From: "Ron G. Minnich" X-Sender: rminnich@terra To: jbryant@tfs.net cc: freebsd-hackers@freebsd.org Subject: Re: Building parallel "Beowulf-style" supercomputers with FreeBSD In-Reply-To: <199710061443.JAA03872@argus.tfs.net> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-freebsd-hackers@freebsd.org X-Loop: FreeBSD.org Precedence: bulk This will be my last reply off-subject, so sorry to those of you who don't care about clusters. On Mon, 6 Oct 1997, Jim Bryant wrote: > Was not the first at Sandia, using VAXen? '88? The first real demo of a 'cluster supercomputer' benched against a supercomputer was done AFAIK by Mary Mock. She competed the Apollo corporate network against a CDC 7600. The Apollos won, once she got about 150 or so into the mix. This was in .... 1984! She won the usual award for those who are ahead of their time: no one ever cites her or acknowledges the contribution she made. Ah well. Tom Nash et. al. at Fermilab invented the concept of 'crates' in 1984 or so also, an idea which would look very familiar to the cluster users of today. Clusters/NOWS/Beowulfs/COPS/'you name it' as an idea are over a dozen years old. The idea was proved out a long time ago too. People keep rediscovering it. A common event in our field. > Hmmmmm.... Parallel Virtual Machines or some name like that? I > recall that on the test apps, it outperformed a X/MP or Y/MP... PVM is not the same as the '88 vax cluster. See the PVM book from MIT Press. > Do the programs have to be hand-designed, or is there a vectorizing > compiler available? Most are hand-rolled. Compilers are available commercially as well. They sometimes work. DEC has one. Also you should see Aaron Mark's masters thesis in this area (marks@sarnoff.com). It describes a Parallel C compiler we targeted to: Cray3/SSS, Processor-in-memory systems, and to the SPARC cluster we had at the supercomputing research center. > Of course, I always thought that this was simply an extension of the > transputer concept, except using standard networks, and standard > computers instead. Nothing wrong with this. Good point. The question was asked: can you build freebsd clusters and get good work done. The answer is: you can, and you can join the club: people have been doing it for years. It works. There's lots of different types of software to support you. You need to create a catchy name: that's the important part. I don't know why this is. ron