The Parallel Barnes-Hut tree code PEPC - Scaling, Periodic Boundaries, and Applications

Decanato - Facoltà di scienze informatiche

Data d'inizio: 19 Ottobre 2011

Data di fine: 20 Ottobre 2011

The Faculty of Informatics is pleased to announce a seminar given by Mathias Winkel

DATE: Wednesday, October 19th, 2011
PLACE: USI Università della Svizzera italiana, room A12, Red building (Via G. Buffi 13)
TIME: 12.30am

Classical molecular dynamics simulations with long-range interactions yield an N-body problem that cannot be solved directly for large particle numbers N, even on recent HPC systems. Consequently, efficient summation schemes, such as the Fast Multipole Method or the Barnes-Hut tree code with their respective scalings O(N) and O(NlogN) are indispensable tools for dealing with such problems.
The efficient parallelization of multipole-based algorithms for the N-body problem is one of the most challenging topics in high performance scientific computing. The emergence of non-local, irregular communication patterns generated by these algorithms can easily create an insurmountable bottleneck on supercomputers with hundreds of thousands of cores. To overcome this obstacle an innovative parallelization strategy for Barnes-Hut tree codes on present and upcoming HPC multi- core architectures has been developed. The scheme, based on a combined MPI-Pthreads approach, permits an efficient overlap of computation and data exchange. It shows efficient scaling across the
294,912 cores of the full IBM Blue Gene/P system JUGENE at Jülich Supercomputing Centre and handles up to 2 billion particles.
For simulating virtually infinite bulk systems, the pure mesh-free Barnes-Hut tree code is not applicable. To overcome the classical Ewald approach, a strategy for extending the tree code to (pseudo)periodic boundaries on the basis FMM formalism has been developed and is presented. As an example application, studies of electronic properties in finite Coulomb systems are discussed.

Mathias Winkel, born in 1984, studied physics at Rostock University, Germany. He graduated in theoretical plasma physics on optical and transport properties of warm dense matter in 2009. After working as a research and lecture assistant, he changed to Jülich Supercomputing Centre (Simulation Laboratory Plasma Physics) for a PhD position. Here, he is concentrating on scalability of the parallel Barnes-Hut tree code PEPC and its application to parallel molecular dynamics simulations in the field of plasma physics.

HOST: Prof. Rolf Krause, Dr. Robert Speck