Particle simulations are used in a wide range of applications including molecular dynamics, astrophysics and fluid dynamics. This minisymposium addresses current research on computational performance in particle simulations with particular regard to parallelization and auto-tuning on recent HPC hardware.
Load balancing is a major research topic and subject of this minisymposium. Due to heterogeneity of hardware and due to potentially great variations in particle numbers per thread/process (heterogeneous particle systems), appropriate load balancing algorithms such as tree-based domain partitioning and implementations thereof are required. Besides, with various (and quite different) hardware systems being driven towards the exascale, the automatic exploration and exploitation of the performance of these systems with unified as well as with hardware-aware particle simulation code is highly relevant to mitigate manual tedious code optimization. Therefore, the minisymposium will further address performance prediction for and automatic tuning of particle simulations.
The speakers will give insight into how load balancing, performance prediction or auto-tuning are used and further developed in their implementations which are given amongst others by the simulation codes ls1 mardyn, which is a molecular dynamics package used in process engineering, and FDPS, a framework for hybrid large-scale parallelization of particle-based simulations.
The minisymposium is embedded into the program of the conference SIAM PP 2018.
|Date||Saturday March 10, 2018|
|Venue||Waseda University, Tokyo, Japan|
The minisymposium is organized by