A Barotropic Solver for High-Resolution Ocean General Circulation Models

High-resolution global ocean general circulation models (OGCMs) play a key role in accurate forecasting. However, the of operational forecasting systems are still not high resolution due to subsequent demand for large computation, as well low parallel efficiency barrier. Good scalability is an important index and challenge OGCMs. We found that communication cost barotropic solver, namely, preconditioned conjugate gradient (PCG) method, bottleneck frequency reductions. In this work, we developed new algorithm—a communication-avoiding Krylov subspace method with PCG (CA-PCG)—to improve then applied it Nucleus European Modelling Ocean (NEMO) example. For PCG, inner product operations were needed every iteration, while CA-PCG, only eight iterations. Therefore, decreased from more than 94.5% total execution time less 63.4% CA-PCG. As result, modes 17,000 s 6000 18,000 6200 Besides, ratio speedup can also be increased 3.7 4.6. summary, process count when using CA-PCG was effectively improved providing highly effective solution simulation.