chapter  10
Preparing the Community Earth System Model for Exascale Computing
ByJohn M. Dennis, Christopher Kerr, Allison H. Baker, Brian Dobbins, Kevin Paul, Richard Mills, Sheri Mickelson, Youngsung Kim, Raghu Kumar
Pages 24

Understanding the complex interactions present in the Earth's physical and biological systems within the ocean, atmosphere, land, and ice has long been of scientific interest. This chapter discusses the preparation for exascale of a widely used and well-respected climate model: the Community Earth System Model (CESM). Earth-system models continue to grow even larger and more complex as they evolve to include new scientific features and take advantage of advances in high-performance computing technologies. The high-fidelity climate simulations that are run at exascale produce massive output data sets and processing these is challenging at best. The single-core performance optimizations implemented included improving vectorization by redesigning data structures, improving data locality by rearranging calculations for cache reuse, minimizing the number of expensive divides, and enhancing compile-time specifications for loop and array bounds. In the dynamics, tracer, and vertical remap modules, the code is blocked over tracers and in the dissipation module the code is blocked over the vertical levels.