A Review of the exploitation of parallel processing in environmental modelling

Publisher Summary This chapter reviews the published achievements in the use of parallel processing for environmental modeling projects. Most oceanographic and shallow water models represent spatial regions using two- or three-dimensional finite difference grids. The discretization of the hydrodynamic equations usually demands only nearest neighbor or at most short range, communications across the grid. When explicit single-step time integration methods are used (the most common), the models are well suited to a highly parallel computer constructed as a mesh of simple processors. Recent experience has shown that present day interprocessor communication rates are not usually a serious limitation to overall performance, enabling good parallel speed-up to be achieved. At each grid point, the ratio of floating point operations to memory accesses is low, and a high memory bandwidth at each node is, therefore, highly desirable. This has been found to limit the single node performance on some parallel machines. At intervals during the calculation, a number of field variables across the whole model need to be saved, usually by writing on disk files. It is essential that parallel systems incorporate a high performance parallel input/output system with the facilities for the visualization of large datasets.