Computational Kernel Algorithms for Fine-Scale, Multiprocess, Longtime Oceanic Simulations

Abstract

Abstract Progress in computer technology has made it possible to make larger calculations with finer grid-scale resolution, and physical processes that were beyond the reach of coarse-resolution models are now simulated directly. This focuses scientific interest toward more turbulent flow regimes and applications toward more realistic modeling of specific regional configurations. In this chapter, we examine the numerical design of oceanic modeling codes specifically suited for modern demands. These are compared with traditional “legacy” oceanic general circulation models and with computational fluid dynamics methods for modern engineering applications. Our primary subject is how the numerical algorithms for different aspects of the discretized partial differential equation system — the computational kernel — combine to yield the overall model performance, with particular focus on avoiding destructive interference among algorithmic components.