Building and Programming a Beowulf Computer

by Travis Kuster
Faculty advisor: Dr. Shawn Stone

Problems in science are increasingly being modeled and solved on computers. These problems are large and on a single computer may take years to run. Before 1994, problems were solved on very expensive supercomputers that lived at government labs or top academic institutions. Since then NASA engineers figured out that you could construct a supercomputer from single computers and network devices into a cluster to solve problems. They called this new type of supercomputer a Beowulf cluster. Currently 414 of the top 500 supercomputers are clusters (TOP 500 Supercomputers List, 2010). Due to the low cost and easy availability of desktops and network hardware, building a powerful Beowulf cluster is now easier than ever. For this research, a nine node Beowulf cluster was built from off the shelf hardware and free software available over the internet. In order to demonstrate the power of a parallel architecture, an n-body simulation written for the cluster is run in serial and parallel form. Results from this and other benchmarking software will be presented as well as information on how to construct your own Beowulf cluster.