Optimizing molecular dynamics simulations with product lines

摘要

This paper presents a case study of using product-lines to address the variability of optimization methods and target platform mappings in high-performance molecular dynamics simulations. We use Feature Oriented Programming to incrementally extend the base algorithm by composing performance enhancement features with the core functionality. Developed features encapsulate common optimization methods in molecular dynamics simulations and target platform mappings. The main benefits of the approach are: 1) it promotes an incremental development, where optimizations and mappings are developed incrementally and simultaneously with the core functionality; 2) complex optimizations and mappings can be obtained by composing basic features. The performance of synthesized products is comparable to the performance of products developed with traditional parallel programming techniques. In this approach complex optimizations become easier to develop, by composing basic features, providing a performance advantage over traditional programming techniques.