INVERSE-DISTANCE-WEIGHTED SPATIAL INTERPOLATION USING PARALLEL SUPERCOMPUTERS

Interpolation is a computationally intensive activity that may require hours of execution time to produce results when large problems are co nsidered. In this paper we describe a strategy to reduce computation t imes through the use of parallel processing. To achieve this goal, a s erial algorithm that performs two-dimensional inverse-distance-weighte d interpolation is decomposed into a form suitable for parallel proces sing in two shared memory computing environments. The first uses a con ventional architecture with a single monolithic memory, while the seco nd uses a hierarchically organized collection of local caches to imple ment a large shared virtual address space. A series of computational e xperiments was conducted in which the number of processors used in par allel is systematically increased. The results show a substantial redu ction in total processing time and speedups that are close to linear w hen the additional processors are used. The general approach described in this paper can be used to improve the performance of other types o f computationally intensive interpolation problems.