Scalability of Semi-implicit Time Integrators for Nonhydrostatic Galerkin-Based Atmospheric Models on Large Scale Clusters

We describe our ongoing effort to optimize a next-generation atmospheric model based on the continuous Galerkin, or spectral element, method: the Nonhydrostatic Unified Model of the Atmosphere (NUMA), which is targeted towards large cluster architectures. The linear solver within a distributed memory paradigm is critical for overall model efficiency. The goal of this work-in-progress is to investigate the scalability of the model for different solvers and determine a set of optimal solvers to be used for different situations. We will describe our novel approach and demonstrate its scalability on a variety of clusters of multicore node clusters. We also present performance statistics to explain the scalability behavior.