Parallel computation on loosely-coupled workstations.

A dominant trend in computing in recent times has been the creation of large networks of computer workstations. We treat these networks as loosely-coupled multiprocessors, and present a model for parallel processing that takes into account the overhead involved in communicating data to and from a remote processor, and derive techniques to optimally partition a given computation. The optimal partition can be determined for a given number of processors and, if required, the optimal number of processors to use can also be derived. Our model can be used to partition computations that can be organized as one- level tree, and are homogeneous and separable. We also present experimental results that validate our model and demonstrate its effectiveness.