A Framework for Parallel Adaptive Grid Simulations

Exploiting parallelism in the solution of scientific and computational engineering problems requires significant expertise and effort on the part of application developers. We describe a framework targeted at the class of discrete-time grid simulations that hides much of the complexity inherent in the parallel solution of those problems. By doing so this programming framework offers the potential to reduce developer training and software development time, and improve the quality of resulting simulation software. Software engineering concerns are important, however, they need not stand in the way of high-performance. Our framework is designed to admit efficient solution to simulation problems and to allow developers to address simulation problems that have a high-degree of dynamism and irregularity. We evaluate both the usability and performance of the framework as applied to a standard finite-difference computation. Keywords : parallel software, parallel discrete event simulation, gri..