Improving Throughput of the ACME Climate Model by Parallel Splitting Atmospheric Physics and Dynamics

Abstract:

If fluid dynamics and atmospheric physics parameterizations were computed in parallel, they could be calculated simultaneously on separate cores of a supercomputer. This would greatly increase model throughput for high-resolution simulations. Additionally, because atmospheric physics is embarrassingly parallel, more sophisticated physics parameterizations could be used without slowing simulations down by simply increasing the number of cores used. The downside to this approach is that it increases time-truncation error. In this presentation, we demonstrate that parallel splitting the ACME model and using a smaller timestep for physics results in faster, more accurate solutions.