Using distributed memory dynamically in a cluster of PVM-based Linux workstations

Parallel virtual machine (PVM) has been identified as an effective system for heterogeneous computing. In this paper we investigate the performance of a heterogeneous distributed system based on PVM, where the computing nodes are connected each other through several levels of communication hierarchy and have different specifications. Matrix computation is taken as sample case for checking the cost and performance of the heterogeneous PC cluster. Experiment shows that data size needs to be as big as possible and communication cost increases with increase in number of hosts for fixed size problem. Heterogeneous PC cluster which is cost effective and performance oriented, can used in providing quality education and research for weather forecasting, image processing and many more areas of scientific computing.