-----Original Message----- From: Adam Martin <[EMAIL PROTECTED]> To: [EMAIL PROTECTED] <[EMAIL PROTECTED]> Date: Thursday, March 28, 2002 5:32 PM Subject: Beowulf Cluster
>Ok, I have been doing a bit of reading up on Beowulf clusters... and I >am now intrigued as to how good they really are... Wayne's rule of thumb: Add up the MHz and multiply by 90%. E.G. 4 x Pentium 1800 = 7200, 7200 x 90% = approx 6470. So four Pentium 1800's run as fast as one Pentium 6470. >1. Does anybody on the list use / or has used one? I haven't used a Beowulf but I have built a couple of clusters of workstations. (Beowulf-HOWTO section three) Here's a cluster the Auckland:University built: > A 214-node DebianGNU/Linux based cluster named Kalaka has been constructed > at the Tamaki Campus, University of Auckland, New Zealand. It is a > distributed system for high performance parallel computing that has so far > been used in several math and physics related projects.> > A web page describing the cluster setup, configuration and software > is availavle at:> <http://www.scitec.auckland.ac.nz/~peter/kalaka.html> >2. What kind of jobs have / can be achieved with such a setup? It all depends. http://www.linuxdoc.org/HOWTO/Beowulf-HOWTO-4.html#ss4.1 section Application Architecture and section 4.4 Suitability. If the application does not have any parts that can be run concurrently then there is no point in running it on more than one processor. I've seen articles on Beowulfs modeling the life of quasars, and overlaying colour and black and white satelite photos of different scales. And rendering graphics. >3. Would anyone be interested in setting up a cluster and doing some >experiments? Maybe. >In particular I am wondering if it would be easy to implement such jobs >as graphical rendering, Done that. There are three basic ways: Beowulf/PVMpov, MOSIX and povray, or a simple cluster of workstations. Details on a simple cluster of workstations on my homepage http://homepages.ihug.co.nz/~wrooney/present > major database queries Not quite sure how you'd set that one up. One copy of the database for each node? Otherwise you're going to bottleneck at the hard drive/bus. You'd probably want a load-sharing cluster for databases rather than a Beowulf. >Looking forward to hearing your responses. You've heard it :-) >Cheers >Adam Wayne
