|Name:||Large Scale Simulations(11) Load Balancing for Large Scale Particle-in-Cell Plasma Simulation in PICADOR|
|Time:||Monday, June 17, 2013
2:15 PM - 2:20 PM
|Room:||Multi-Purpose Area 4 (MPA 4)
CCL - Congress Center Leipzig
|Speakers:||Sergei Bastrakov, UNN|
|Abstract:||One of the currently high-demand areas of computational physics is simulation of plasma dynamics with the Particle-in-Cell method. Solving up-to-date physical problems requires large scale plasma simulation on a cluster system with thousands of cores. Therefore, there is an interest in techniques of high-performance implementation of the method for both traditional cluster systems and, particularly, heterogeneous systems equipped with GPUs.
PICADOR is a tool for three-dimensional plasma simulation based on the Particle-in-Cell method. It is capable of using CPUs and GPUs of a heterogeneous cluster system and scales up to at least 2048 CPU cores and 512 GPUs with weak scaling efficiency of 90% and 64%, respectively, on naturally well-balanced tasks. For tasks with highly non-uniform particle distribution load balancing is essential.
We present recent developments in implementing static and dynamic load balancing in PICADOR. The load balancing strategy is based on rectilinear partitioning scheme. On moderately imbalanced tasks we observe, on average, 30% performance and some scaling efficiency improvement.
Sergei Bastrakov, UNN; Evgeny Efimenko, Russian Academy of Sciences; Arkady Gonoskov, Russian Academy of Sciences; Alexander Malyshev, UNN; Iosif Meyerov, UNN; Igor Surmin, UNN