| Summary: | 2018 Fall. Includes bibliographical references. Earth System models (ESMs) are complex software packages comprised of millions of lines of code used to simulate many different Earth processes. ESMs simulate the dynamical and physical states of the atmosphere, land, ocean, sea ice, rivers, and glaciers and coordinate the interactions between them. Many computational challenges exist within these models and future systems are putting more pressure on these challenges. In order to alleviate some of the pressure, it is important to study the performance challenges that exist within the models in order to understand the optimizations that need to be performed as we move to exascale systems. This work studies the performance of the coupling infrastructure between the modeling components within the Community Earth System Model. The coupler is responsible for the data exchanges between the different modeling components and while it has a small computational footprint, it has the potential to have a large impact on performance if the component resources are dispersed in incorrect proportions. This work explains and addresses this issue and provides easy solutions for users to save thousands of core cpu hours.
|
|---|