Parallel Programming Languages: Do They Meet the Needs of Scientists and Engineers?

Although computer scientists use parallel programming languages to model concurrency, scientists and engineers approach parallelism with a different objective. The physical world they model is inherently parallel, but scientific programmers have become accustomed to using sequential techniques for its study. It is the ability to handle massive amounts of data that motivates their conversion to parallelism. This report considers the current state of the art of parallel programming languages from the perspective of scientists and engineers. It reviews the problem-solving strategies commonly employed by scientific programmers and compares them with the conceptual models for parallelism supported by high-level programming languages available to the general user community. The fact that existing languages were designed and implemented by members of the computer science profession is reflected in the prevalence of features representing that community''s view of parallelism. Consequently, scientific programmers are unlikely to find language models that map cleanly to their problem-solving approaches.