With exascale computer systems, an unprecedented level of detail in the description of our Earth can be achieved, with the hope of significantly reducing previously persistent uncertainties in climate projections. However, the massive parallelism and large heterogeneity of exascale computers pose a particular challenge to the efficient execution of Earth system models. In the project "Optimization of simulation algorithms for exascale systems for the computation of the Earth system model ICON", novel methods of parallel execution, communication and dynamic load balancing for the Earth system model ICON will be implemented and optimized. Existing concepts of parallel execution, which are based on a spatial decomposition of the computational domain, will be extended by functional concurrency, and the resulting additional data exchange will be optimized in a co-design approach. Furthermore, a uniform load distribution is ensured for the model components by regularly optimizing the spatial decomposition during runtime through dynamic load balancing. In addition, the potential of hardware accelerators is tapped in a portable manner. These technological advancements will be tested in the context of two outstanding scientific use cases: the formation of clouds and precipitation in the atmosphere and the biogeochemistry of the ocean.




1 full time equivalent

Funded by: