MPI Parallelisation of 3D Multiphase Smoothed Particle Hydrodynamics
Xiangda Cui, Wagdi G. Habashi, Vincent Casseau
Abstract
This paper addresses the MPI parallelisation of a 3D multiphase smoothed particle hydrodynamics (SPH) solver, focusing on supercooled large droplets (SLD) impingement. SPH uses moving particles to represent fluid flows, demanding particular parallelisation strategies. A cell system provides a spatial reference to generate dynamical neighbour lists, and to pack, send and receive particles. A re-indexing method is introduced to accommodate migrating particles and a sequential particle collecting technique is proposed to reduce the complexity of communications for 3D partitioning. The scalability is tested on up to 1024 processors with 110 million particles, and the effects of particle load on computing speeds are also investigated. The solver is then applied to droplet impingement at various impact speeds and diameters to study the consequent splashing. This work provides a generic approach for highly multi-threaded SPH solvers, allowing massive computations for improved accuracy and/or numerous runs for parametric studies.