Parallelization of a dynamic unstructured algorithm using three leading programming paradigms

The success of parallel computing in solving real-life computationally inte nsive problems relies on their efficient mapping and execution on large-sca le multiprocessor architectures. Many important applications are both unstr uctured and dynamic in nature, making their efficient parallel implementati on a daunting task. This paper presents the parallelization of a dynamic un structured mesh adaptation algorithm using three popular programming paradi gms on three leading supercomputers. We examine an MPI message-passing impl ementation on the Gray T3E and the SGI Origin2000, a shared-memory implemen tation using the cache coherent nonuniform memory access (CC-NUMA) feature of the Origin2000, and a multithreaded version on the newly released Tera M ultithreaded Architecture (MTA). We compare several critical factors of thi s parallel code development, including runtime, scalability, programmabilit y, portability, and memory overhead. Our overall results demonstrate that m ultithreaded systems offer tremendous potential for quickly and efficiently solving some of the most challenging real-life problems on parallel comput ers.