|Petascale Global Kinetic Simulations of The Magnetosphere and Visualization Strategies for
Analysis of Very Large Multi-Variate Data Sets
|444, 5th International Conference of Numerical Modeling of Space Plasma Flows (ASTRONUM 2010)
|Karimabadi, H.; Loring, B.; Vu, H. X.; Omelchenko, Y.; Tatineni, M.; Majumdar, A.; Ayachit, U.; Geveci, B.
|3D global electromagnetic hybrid (fluid electrons, kinetic ions) simulations have long been
considered the holy grail in kinetic modeling of the magnetosphere but high computational
requirements have kept them out of reach. Petascale computers provide the computational
power to make such simulations possible but peta computing poses two technical challenges.
One is related to the development of efficient and scalable algorithms that can take advantage
of the large number of cores. The second is related to knowledge extraction from the resulting
simulation output. The challenge of science discovery from the extremely large data sets (∼ 200
TB from a single run) generated from global kinetic simulations is compounded by the multi-variate
and “noisy” nature of the data. Here, we review our innovations to overcome both challenges.
We have developed a highly scalable hybrid simulation code (H3D) that we used to perform the
first petascale global kinetic simulation of the magnetosphere using 98,304 cores on the NSF
Kraken supercomputer. To facilitate analysis of data from such runs, we have developed
complex visualization pipeline including physics based algorithms to detect and track
events of interest in the data. The effectiveness of this approach is illustrated through examples.