The research field aims at understanding the physical processes of large scale geological incidents (e.g. earthquakes) and their secondary effects (e.g. tsunamis).
This is essential, first, for allowing rapid response during and after major catastrophic events, and second, for improving technical and social readiness. Numerical solutions to the governing non-linear partial differential equations play a key role in linking the underlying physics with observational data. The wide range of involved scales in space and time as well as the non-linear nature that underlie these natural phenomena pose an enormous computational challenge that can be tackled only by large-scale HPC-resources.
Developments within DEEP-ER
Within the project, the compute kernels as well as the I/O-strategy of a state-of-the-art parallel finite-element research application is optimised for the Intel MIC and the DEEP/-ER Architecture. The efforts undertaken include:
- optimisation and vectorisation of the compute kernels (e.g. sparse-matrix multiplications)
- improvement of large-scale I/O through suitable I/O-abstraction layers
- enabling efficient checkpointing and simulation-restart
- enhancement of visualisation and data-mining capability of large-scale simulation-database, through efficient data formats and compatible external database-tools and/or scientific gateways