Developing molecular dynamics simulation codes using mixed language programming

摘要

We describe our experiences parallelizing a large-scale scientific application to model systems of discrete particles. We describe the approach and tasks undertaken to parallelize this application using two different programming paradigms: imperative and functional. The objectives of both exercises were to maximize performance, parallelism and portability, and to minimize program development costs. We believe this study reveals an important relationship between conventional and novel parallel programming paradigms, and identifies important attributes that novel paradigms must have to gain wide acceptance.