SEP has always been interested in problems that stretch our computational resources. Over time the high-end computer at SEP has gone from array processors Newkirk (1977) to the Convex Claerbout (1985) to the CM5 Biondi (1991) to the Power Challenge to multi-processor Linux machines Biondi et al. (1999), and in the last year, its first Linux Beowulf cluster.
This fall SEP decided to further expand its computer resources. We found the most cost effective option was to build our own Beowulf cluster. This paper attempts to summarize our cluster building experience and to explain how we program and manage our new cluster. In addition we discuss the current limitation of our setup and future plans to overcome some of these obstacles.