Unstructured Adaptive Grid Computations on an Array of SMPs

Dynamic load balancing is necessary for the parallel adaptive solution of unsteady problems in fluid dynamics, since their computational requirements change as the simulation progresses leading to load imbalance. JOVE is such adynamic load-balancing framework. We study the performance of two different implementations of JOVE on the Silicon Graphics' POWER CHALLENGEarray. This parallel machine is an array of shared-memory symmetric multiprocessing (SMP) systems, an architecture that is becoming increasingly popular as the most useful model of scalable parallel computing. Parallel algorithms need to be designed to exploit the hybrid communication model offered by such an architecture, and in this paper, we study these issues as they relate to JOVE. 1