Software infrastructure development on the LUMI supercomputer to deploy the next generation Earth system models

Abstract

The Destination Earth (DestinE) climate adaptation digital twin (climate DT) will be run on the EuroHPC supercomputers LUMI and Marenostrum. The climate DT project brings together leading centers from across Europe specializing in climate science and services, Earth system modelling, and supercomputing to deliver an innovative climate information system through DestinE supporting the European Union’s climate adaptation efforts. Usage of the Earth-system model simulations requires an orchestration of a number of software tools that are developed and implemented as part of the climate DT project. Here, we present the progress of the software infrastructure development and support on LUMI to deploy the next generation Earth system models ICON (Icosahedral Nonhydrostatic) and IFS (Integrated Forecasting System) coupled with the ocean models NEMO (Nucleus for European Modelling of the Ocean) and FESOM2 (Finite-volumE Sea ice-Ocean Model). Our priority is to provide the computing environment needed on LUMI for (i) the development and execution of the climate models, impact models and data analysis utilities, (ii) the infrastructure components needed in terms of data storage, discovery, and quality control both internally and towards a link with the DestinE data services, and (iii) the software development tools and optimization support during the development of the climate models to ensure LUMI is efficiently used to reach the desired throughput. Several software tools have been deployed on LUMI successfully, for instance, MultIO the I/O server to execute the climate models outputs, FDB (Fields Data Base) for the data storage and access, Autosubmit and ecFlow for orchestration of workflows on LUMI which configures, executes, and monitors the climate DT experiments. Further, GPU adaptation of ICON and IFS-NEMO/FESOM2 models on LUMI-G (GPU) is undertaken. At the moment IFS and ICON compile and run successfully on LUMI-C (CPU). Profiling and performance analysis of the Earth system models (IFS-NEMO/FESOM2 on LUMI-C and ICON on LUMI-G) will be presented with throughput and performance numbers for the baseline test case configurations. Issues and challenges will be discussed in detail in the presentation.