Parallel processing in the networks of workstations

Abstract

A model and statistical algorithms for utilizing networks of workstations as distributed parallel computing systems are described. Each workstation in this model consists of two processors: one for computation and one for communication. The advantages of this model are that the workstations can assign the intensive load balancing and the communication jobs to the communication processor and the code execution (including compiling and linking) jobs to the computation processor. The communication processor provides a quick response to the user and/or to the requests from the other nodes, and the computation processor provides a high throughput without interprocessor communication interrupt.