A Survey on Parallelism and Determinism

Parallelism is often required for performance. In these situations an excess of non-determinism harmful as it means the program can have several different behaviours or even results. Even in domains such high-performance computing where parallelism crucial performance, computed value should be deterministic. Unfortunately, programs also allows dynamic scheduling tasks, reacting to first task that succeeds, cancelling tasks cannot lead a result, and so on. Non-determinism thus both desired asset undesired property depending on situation. practice, necessary limit identify precisely sources control what parts are deterministic not. This survey takes perspective programming languages, studies how models ensure determinism parallel programs. not only languages but prevent one particularly demanding source non-determinism: data races. Our objective compare existing solutions following questions: How help programmers write run manner without visible non-determinism? What paradigms this kind properties? We study questions discuss merits limitations approaches.