BUILDING PARALLEL COMPUTERS WITH MINIMAL COSTS FOR HIGH PERFORMANCE COMPUTING COURSES

Since 2003, the speed of uniprocessors can hardly be further pushed because of energy-consumption and heat-dissipation problems. That is why the major processor manufacturers like Intel and AMD introduce multi-core processors instead of faster uniprocessors since then. Consequently, all cores of a multi-core processor have to be utilized in order to improve the performance of the system. Parallel programs have to be coded and executed to take advantage of all cores of a multi-core processor. Both phenomena push the need for Hihg Performance Computing (HPC) education which is based on parallel processing to be included in contemporary computer science curriculum. However, the largest barrier for an institution, especially a small college, to offer HPC courses is the high cost of high performance computers. While HPC education is imperative, very few institutions can afford expensive parallel computers. In order to overcome the major obstacle, the author presents in this paper how to build an affordable parallel platform, namely a commodity cluster, with free open source software and retired personal computers (PC’s) so that any institution can construct parallel computers for its HPC courses for minimal costs.