Global Atmospheric and Ocean Modeling on The Connection Machine

Abstract

This paper describes the high-level architecture of two parallel global climate models: an atmospheric model based on the Geophysical Fluid Dynamics Laboratory (GFDL) SKYHI model, and an ocean model descended from the Bryan-Cox-Semtner ocean general circulation model. These parallel models are being developed as part of a long-term research collaboration between Los Alamos National Laboratory (LANL) and the GFDL. The goal of this collaboration is to develop parallel global climate models which are modular in structure, portable across a wide variety of machine architectures and programming paradigms, and provide an appropriate starting point for a fully coupled model. Several design considerations have emerged as central to achieving these goals. These include the expression of the models in terms of mathematical primitives such as stencil operators, to facilitate performance optimization on different computational platforms; the isolation of communication from computation to allow flexible implementation of a single code under message-passing or data parallel programming paradigms; and judicious memory management to achieve modularity without memory explosion costs.