RHS - A FRAMEWORK FOR EXPLOITING DISTRIBUTED PARALLELISM EFFICIENTLY

The proposed reciprocal hypercomputer system (RHS) framework supports distributed computing by providing three important elements; reciproca l software execution, load balancing and local compilation. Together t hese elements provide an efficient way to exploit large-grain parallel ism. All currently known hypercomputer systems are experimental. Not o ne of them has established predominance or has explicitly combined the three elements in a single framework. The C_APPL language is used to explain parallel programming in the RHS, implemented on a local area n etwork of SPARC workstations and personal computers. Results from four experiments are presented, namely fractal calculation, the travelling salesman algorithm, lengths of all pairs' shortest paths, and matrix multiplication. The RHS results are compared with the p4 results gener ated on the same platform for the travelling salesman algorithm. Our e xperience suggests that the RHS can solve real-life problems efficient ly.