Parallel computation with dynamic load distribution for locally distributed Windows NT environments

Abstract Today 's requirements for computational power are still not satisfied. One answer for this demand is expensive supercomputers. Another attempt is the collection of computational power in a network. Connected workstations operate for parallel computation. Our approach for collecting unused resources in workstation clusters enables dynamic distribution of computational load over the network. Any computer individually distinguishes the amount of server resources sharing for others in the network. Our architecture is based on the client /server model —clients divide potentially hard problems into sub problems and forward them to different servers in the network. The usage of standardized remote procedure calls (rpc) as the basic mechanism for transmission of data between the workstations allows the extension of the concept for heterogeneous environments. This article gives a discussion of our implementation for Windows NT and presents the latest benchmarks with two special parallel applications.