Author:

Keywords:

Science & Technology, Technology, Computer Science, Hardware & Architecture, Computer Science, Vortex particle method, remeshing, parallel Barnes-Hut tree code, parallel sorting, IBM Blue Gene/P

Abstract:

Parallel vortex particle methods are an efficient technique for massively parallel simulations of turbulent fluid flows. One of their big advantages is the intrinsic adaptivity of vortex particles, since computational elements exist only where the vorticity field is non-zero. To overcome O(N2)- complexity of the corresponding N-body problem, multipole-based fast summation methods can be used. However, the convergence condition of vortex particle methods is only satisfied for very short times, prohibiting long-term simulations. To circumvent this, many recent codes use the concept of remeshing with an underlying mesh structure. In this paper, we demonstrate that the classical remeshing technique can be implemented directly and efficiently into a mesh-free parallel Barnes-Hut tree code. Using a dynamic 3D numerical example, we analyze the scaling behavior of this algorithm from 512 to 16,384 cores on an IBM Blue Gene/P system. © 2012 The authors and IOS Press. All rights reserved.