文摘
This paper presents a novel parallel design for Smoothed Particle Hydrodynamics. The Message Passing Interface (MPI), standard for distributed memory programming, is used to parallelize the code as a necessary precursor to future multi–GPU implementation. In the proposed parallelization scheme, the domain decomposition is performed based on both spatial and particle decompositions to reach efficient and well-balanced parallelization. To take the advantage of memory locality, the Peano–Hilbert ordering of the underlaying cells, which allows particles that are spatially close to also be close in memory, is adopted. In our scheme, the dynamic load balancing is performed in three Cartesian dimensions as a feedback system that recognizes the particle imbalance and applies the load balancing accordingly. The incompressible SPH (ISPH) method along with an eddy viscosity turbulence model is solved explicitly in the proposed parallel scheme, rather than the typical weakly compressible or implicit incompressible Poisson formulations. The performance of the code is tested for several test cases including a dam–break problem impacting on a short box. The simulation results for water depth at two locations in the tank and two pressure sensors on the box are compared with experimental data and reasonable agreement is achieved.