Towards exascale simulations of space plasmas using the DEEP-ER architecture.


Space weather, the science of studying and forecasting the conditions in the ionised gas (plasma) in between the Earth and the Sun, has a high societal impact.

The impact lies in its relevance to the space and ground infrastructure and to the external natural drivers of the Earth climate.


KULeuven has developed a first principle approach implemented in the code iPic3D and based on modelling a statistical sample of electrons and ions and the electromagnetic fields of the space environment. iPic3D is chiefly suitable for the DEEP/-ER Architecture, where the highly scalable and highly vectorisable particle operations run in the Booster while the more communication intensive field operations run concurrently on the Cluster.


Developments in DEEP-ER

The relative conceptual simplicity of the key parts of the iPic3D algorithms enable synergistic co-design opportunities where modifications of the algorithm can be easily implemented to test innovative and even radical resilience solutions. Currently both I/O and checkpointing are handled at the application level using HDF5 files saved on the disk by each core at cycle intervals set by the user. The iPic3D application tests the new approaches defined in DEEP-ER in a peculiarly demanding limit: particles and fields reside in different types of cores (Booster and Cluster) and have a vastly different density with hundreds of particles per each field cell. Additionally the particle data structures are size-varying over time as particles move across the domain and are moved across the cores leading to possible imbalances and array sizes that are time dependent and spatially inhomogeneous. We test the approaches for I/O and checkpointing for conditions not seen in other applications.