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.

中文翻译：

---

3D 多相平滑粒子流体动力学的 MPI 并行化

本文讨论了 3D 多相平滑粒子流体动力学 (SPH) 求解器的 MPI 并行化，重点是过冷大液滴 (SLD) 撞击。SPH 使用移动粒子来表示流体流动，需要特定的并行化策略。单元系统提供空间参考以生成动态邻居列表，以及打包、发送和接收粒子。引入了重新索引方法以适应迁移粒子，并提出了顺序粒子收集技术以降低 3D 分区通信的复杂性。可扩展性在多达 1024 个具有 1.1 亿个粒子的处理器上进行了测试，并且还研究了粒子负载对计算速度的影响。然后将求解器应用于不同冲击速度和直径下的液滴撞击，以研究随后的飞溅。这项工作为高度多线程 SPH 求解器提供了一种通用方法，允许进行大量计算以提高精度和/或多次运行参数研究。