Parallelization of a nearshore wind wave model for distributed memory architectures

摘要

Efficient parallel algorithms are required to calculate spectra of random short-crested, wind-generated waves in coastal regions using the third-generation wave model SWAN (Simulating WAves Nearshore). The numerical propagation schemes used in SWAN are fully implicit and appear to be very effective on a sequential computer. On distributed memory parallel computers, however, the inherent sequentiality of these implicit schemes precludes efficient implementation. Using block Jacobi technique seems a good way to improve the parallelism, but it turns out that this comes with a penalty in the form of tripling the number of iterations. An alternative is the block wavefront approach which has the same behaviour as the sequential algorithm with respect to accuracy and convergence. Timing experiments with this approach, obtained on a Beowulf cluster using MPI, reveal good run time reduction.