From chemistry-request ":at:" ccl.net Mon Nov 3 22:43:32 2003 Received: from einstein.physics.drexel.edu (einstein.physics.drexel.edu [129.25.7.60]) by server.ccl.net (8.12.8/8.12.8) with ESMTP id hA43h0hP023578 for ; Mon, 3 Nov 2003 22:43:00 -0500 Received: from si (silicon.physics.drexel.edu [129.25.8.33]) by einstein.physics.drexel.edu (8.11.6/8.11.6) with ESMTP id hA43aeK05693 for ; Mon, 3 Nov 2003 22:36:40 -0500 Date: Mon, 3 Nov 2003 22:42:58 -0500 (EST) From: Avijit Ghosh To: chemistry)at(ccl.net Subject: Re: CCL:Linux Cluster In-Reply-To: <3FA6047B.4070200)at(chem.vu.nl> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Spam-Status: No, hits=-0.5 required=7.0 tests=IN_REP_TO,USER_AGENT_PINE version=2.55 X-Spam-Checker-Version: SpamAssassin 2.55 (1.174.2.19-2003-05-19-exp) We just bought our cluster from a vendor (Western Scientific). I would recommend going to a vendor if you can afford it, mostly to guarantee there won't be any hardware incompatibilities that are sometimes hard to work out until you start setting everything up. Hardware Linux vendors will also quite possibly be more up to date on what the best configurations/tools might be. Price it out and take a look at the relative costs of going the "do it yourself" route and from a vendor. (In our case, our prices were pretty competitive w/ buying cheap 1U dual xeon blades from bulk hardware companies like www.mwave.com and the support is absolutely brilliant). If it looks like it is much cheaper going the do it yourself route, take a look at the following: http://oscar.openclustergroup.org/tiki-index.php Oscar will automate installation over a network and make it reasonably straightforward to install over an arbitrary number of nodes, setup a reasonable queueing system, install mpi(ch), take care of user accounts and maintain patches/updates. The documentation provides some good advice on "reasonable" configurations as well. Read over the "quick install" to get a sense of the work involved for setting up a cluster on your own. Its quite impressive how straightforward this stuff has become. There is no 6 node limit :) best regards, -avi Asst. Prof. Avijit Ghosh, Ph.D. Dept of Physics Drexel University Philadelphia, PA 19104