Object-Oriented Parallel Processing in Distributed Computing Environments

Modern distributed systems consisting of powerful workstations and high-speed interconnection networks represent an economical alternative to special-purpose supercomputers for solving a large number of computationally intensive military and civilian applications. However, the development of parallel processing application software to exploit the potential of such distributed systems remains a formidable challenge. Currently, in developing such applications, the programmer has to contend with the difficult issues of data partitioning, load balancing and fault tolerance that are associated with distributed computing environments. Developing parallel programs for distributed environments is therefore a costly and time consuming task that requires a programmer to have expertise not only in distributed and parallel computing, but also in fault tolerance. This SBIR effort proposes a novel approach for implementing parallel applications on distributed systems that liberates the programmer from the complexities of data partitioning, load balancing and fault tolerance. The proposed approach uses an object-oriented methodology for rapid and cost-effective implementation of parallel processing applications on distributed workstations. With minimal effort, programmers can implement parallel processing applications that are automatically distributed over a heterogeneous network, that are dynamically load balanced as the applications run, and that survive failures of processors and network components in the distributed system.