Large scale complex numerical simulations on large scale complex computer facilities - identifying performance and scalability issues
Enabling studies of real-life physical and engineering problems, based on computer simulations entails growing demand on computational power - demand, driven by increasing space and time scales, and complexity of the scientific problems to be solved. Meanwhile the available computer systems also grow larger and more powerful on the cost of increasing architectural complexity and heterogeneity. The task to match the complex hardware with the algorithms used to simulate problems of multiscale hierarchical nature, and aiming to utilize the computer power to full extend, is by far nontrivial. It questions all decisions along the simulation chain - choosing numerical methods, implementing those, deciding what to optimize and how much. This project addresses the efficiency of the computer simulations in a holistic way. We use a simulation software for a large scale problem from Geophysics/ Quantum Chemistry, developed in Uppsala and analyse its performance together with utilization of the resources via a system analysis tool, provided by the Russian partner. The experiments are utilizing best known numerical algorithms implemented via optimized libraries, to be done on a supercomputer facility that reflects current trends in hardware development. The effort is well aligned with several international activities for achieving excellence in HPC applications.