From owner-freebsd-chat Thu Feb 7 10:12:38 2002 Delivered-To: freebsd-chat@freebsd.org Received: from mail007.syd.optusnet.com.au (mail007.syd.optusnet.com.au [203.2.75.231]) by hub.freebsd.org (Postfix) with ESMTP id C6CE237B423 for ; Thu, 7 Feb 2002 10:12:31 -0800 (PST) Received: from optusnet.com.au (golax6-173.dialup.optusnet.com.au [211.28.134.173]) by mail007.syd.optusnet.com.au (8.11.1/8.11.1) with ESMTP id g17ICSm30413 for ; Fri, 8 Feb 2002 05:12:28 +1100 Message-ID: <3C62C44A.D63FB405@optusnet.com.au> Date: Fri, 08 Feb 2002 04:15:38 +1000 From: Ian Pulsford X-Mailer: Mozilla 4.76 [en] (X11; U; FreeBSD 4.5-STABLE i386) X-Accept-Language: en MIME-Version: 1.0 To: freebsd-chat@FreeBSD.ORG Subject: Re: diskless freebsd cluster References: Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Sender: owner-freebsd-chat@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.org Matt Wilbur wrote: > > Hi all, > > I just wanted to say, to the entire FreeBSD team, especially everyone > involved with pxe support, rc.diskless[1,2], mfs, and gigabit nic support, > THANKS! You ROCK! > > I just got a 16 node diskless cluster of 1.4GHz Athlons (dual 1GHz p3 > server) up and running 4.5-RELEASE, and it was SO much easier than it used > to be.. > > The cluster has two "primary" uses, one is to do high volume number > crunching, the other is for a parallelized port of one of our > toolkits for a particular customer, who is insisting on a Scyld Linux > "beowulf". The original plan was to port our crunching codes from FreeBSD > to Linux (sigh), but since it took about a day to build a second system > disk running FreeBSD (and a port would've taken at least a week or two), > that is far more cost effective in the short. Hopefully I can build a > mini scyld cluster for development only and we can stick with FreeBSD.. > > Why diskless? We're doing diskless because the codes we're currently > using are CPU intensive, with little file i/o, and because the lab it's > operating in has restricted access, its far easier to operate diskless for > media accounting reasons, especially if/when N goes from 16 to say, > 64..or if we want to wheel the rack out of the lab and operate > "outside". I'm getting 97% user CPU on all nodes when I hammer the system > with runs, I can live with that.. The network is 100baseT to the client > nodes, 1000baseSX server->switch. > > For what it's worth, the crunching used to happen on Origin 200s. Of two > primary codes we run, code A would take about the same time on freebsd on > a p3-1GHz as it did on a 270MHz r12000. Code B ran three times faster > on a p3-1GHz than on the 270MHz r12k. Our cost savings are phenomenal > using commodity hardware and a great OS. > > If anyone would find the setup/configuration of interest I can document > how I set it up and throw it up somewhere for your perusal.. It would be neat if you could turn this into a Daemon news article. (BTW, I'm not the one to contact.) Ian To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-chat" in the body of the message