SIParCS 2019 - Lolita Mannik
Novel Database and Usage Analytics for the CESM2 Climate Model: First Steps to Tracking Configuration and Performance
The Community Earth System Model 2 (CESM2) is a highly configurable and complex coupled climate model used by climate researchers across the world. Model configurability comes in the form of multiple components, grids, and physics options, all of which in turn affect the model’s computational performance – an important constraint in designing both model software and scientific experiments. In this study, a database was created by parsing text-based timing files of all experiments run on the Cheyenne supercomputer for the Coupled Model Intercomparison Project 6 (CMIP6) to learn how climate scientists use this sophisticated model. This study identified which configurations are being used, their relative rates of usage, and their computational performance, which can vary significantly. For example, model cost and CPU hours were analyzed by configuration and case, and the effect of a system upgrade on model cost was assessed for those cases running both before and after the upgrade. These performance statistics can also be supplied to the Computational Performance Model Intercomparison Project (CP-MIP) that requests performance statistics for all participating models. This analysis can aid climate scientists in making the best use of their limited computing allocation in relation to the trade-offs between using the latest version at the highest resolution or choosing less sophisticated options due to high computational expense, and it allows NCAR to track configurations over time and understand version adoption rates with the goal of expanding the database to include timing data from all worldwide CESM users.
Mentors: Brian Dobbins, John Dennis