Using a hybrid platform for cluster, NoW and GRID computing

Clusters, Networks of Workstations (NoW) and Grids offer a new, highly available and cost-effective parallel computing paradigm. Their simplicity, versatility and unlimited power made them a rapidly adopted technology. It seems that they form the new way of computing. Although these platforms are based on the same principles, they differ significantly, needing special attention to their characteristics. Future computer scientists, programmers and analysts have to be well prepared, both about administrative and programming issues, in order to ensure faster and smoother transition. We present the architecture of a dynamic clustering system consisting of Beowulf class clusters and NoW, as well as our experience from constructing and using such a system, as an educational infrastructure for HPC. nema