An efficient parallel model for coastal transport process simulation

A three-dimensional (3D) parallel model for efficient simulation of sediment-water transport processes in coastal regions is introduced in this paper with a main focus on the parallel architecture of the model. The model's parallel efficiency is maximized in two steps. First, a fully parallelizable hybrid operator splitting numerical technique is applied to discretize the governing equations of the model. Within one computation time step, the technique makes use of an explicit Eulerian-Lagrangian scheme to deal with the advection, and a finite element method and a finite difference method to handle the horizontal and vertical diffusion, respectively. Second, the model parallelization is established according to a mixed strip/patch domain decomposition method and master-slave communication approach on multiple instruction multiple data (MIMD) machines. In order to test the model efficiency and accuracy, two real-life applications were conducted on a MIMD parallel computer, IBM SP2/900. The computed results agree reasonably well with field measurements. A new iterative matrix solving technique developed especially for coastal process calculations, namely the Gauss-Jacobi (GJ) iteration method, has also been implemented in the two applications. With the new technique, the performance of the parallel model can be improved five-fold for each extra processor used. The performance leap greatly increases the engineering relevance of the model, for instance, predicting the long-term fate of dredged material disposed in a dynamic environment.