Final Report on Porting Earth System Models to Containers - Deliverable D2.9
Authors/Creators
- 1. Swiss National Supercomputing Centre, ETH Zurich
- 2. Center for Climate System Modeling, ETH Zurich
Description
Atmosphere and ocean models are characterised among other things by complex dependencies, external configurations, and performance requirements. The containerisation of such a software stack helps to provide a consistent environment to ensure security, portability and performance. Since the container is built only once but then can be deployed on multiple platforms, productivity is increased.
ETH Zurich has led a multi-year effort to containerise key models from the climate / numerical weather prediction community. This effort started in 2018 with a “hackathon” (programming session) to help ESiWACE2 community scientists create containers for their Earth-system models. The six teams made fast progress on their containers and their results were reported in the ESiWACE2 deliverable “D2.8 Summary of Container Hackathon Experiences”. Of these teams, three received extended ESiWACE2 funding to refine their containers : LFRic, EC-Earth, and COSMO. Even though they did not receive funding, the ICON team viewed their containerisation as strategic and has offered their results as part of this final deliverable. Therefore, we report here on four model containers – LFRic, EC-Earth, COSMO and ICON – rather than the originally foreseen three.
The timing of the hackathon – just before the Covid-19 pandemic – was fortunate, for the teams were able to return home to refine their containers and put them into a production framework with only virtual interaction. ETH Zurich provided technical assistance to the teams, which became highly independent with time. The requirement for the final deliverable was not only the containerisation itself but also that the resulting model performance did not suffer adversely. The overall process and an overview of the containerization work is given in the subsequent sections, each of which references the public online documentation which the teams developed during their efforts.