Hybrid Cell Finite Volume Euler Solutions of Flow Around a Main-Jib Sail Using an IBM SP2

Abstract

This chapter portrays the parallel implementation of an explicit Euler solver, developed to allow the use of any cell topology using customized data structures. While near ideal speed-ups demonstrate the efficiency of the parallel algorithm on an IBM-SP2, the utilization of these data structures for adaptive gridding and dynamic load balancing is described. An approach, independent of cell topology, has been developed using customized data structures. The Finite Volume method has been utilized to implement a scheme capable of explicitly solving the Euler equations over unstructured hybrid meshes constructed from any discrete cell topology. The Cell Vertex method employed is easily implemented, while the approach separates the grid generation method from the flow solver. The IBM-SP2 based at the University of Southampton provides an ideal platform for such a code. Distributed memory architectures provide substantial memory resources, necessary for the high cost of utilizing unstructured data types, while efficient communication between processors allows high speed-up rates to be achieved. The Finite Volume Euler scheme has been developed using a cell vertex explicit upwind scheme to calculate fluxes on Control Volume faces. Efficient implementation of used data structures allows easy, efficient manipulation of the data providing the potential use of adaptive gridding and dynamic load balancing. The algorithm has been parallelized using MPI and ported onto an IBM-SP2 at the University of Southampton. The parallel implementation has been achieved easily utilizing the versatile data structures to construct communication lists across the processor array.