Implementation of SPH and DEM for a PEZY-SC Heterogeneous Many-Core System
Implementation of SPH and DEM for a PEZY-SC Heterogeneous Many-Core System
复制标题
PEZY-SC异构众核系统的SPH和DEM实现
DOI：
10.1007/978-3-030-27053-7_60
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
Furuichi Mikito
中科院分区：
文献类型：
--
作者：
Hosono Natsuki;Furuichi Mikito
基于粒子的模拟策略，如光滑粒子流体动力学（SPH）和离散元方法（DEM）的方法是有用的工业和防灾应用。一个成功的大规模粒子模拟的关键问题之一是能源（电力）成本。为了缓解这个问题，使用PEZY-SC众核微处理器应该提供一个有效的解决方案，因为它们的功率效率很好。我们已经实现了并行SPH和DEM代码的基础上开发粒子模拟器（FDPS）。我们的策略涉及到使用PEZY-SC微处理器作为一个外部加速器设备在一个卸载型的实现，其中的交互内核被分配给这些微处理器和前/后处理主机CPU上执行。性能测试表明，计算交互内核的线程数具有良好的可扩展性，但计算成本的细分显示了进一步改进的空间。水和颗粒溃坝试验也进行了PEZY-SC系统，以验证SPH和DEM代码，分别。
Particle-based simulation strategies such as the smoothed particle hydrodynamics (SPH) and discrete-element method (DEM) approaches are useful in industry and disaster-prevention applications. One of the key problems for a successful large-scale particle-based simulation is energy (electricity) costs. To mitigate this problem, using PEZY-SC many-core microprocessors should provide an effective solution because their power efficiency is good. We have implemented parallelized SPH and DEM codes based on Framework for Developing Particle Simulator (FDPS). Our strategy involves using the PEZY-SC microprocessors as an external accelerator device in an off-load-type implementation in which the interaction kernel is allocated to these microprocessors and a pre/post-processing is performed on the host CPU. Performance tests show good scalability of the computing interaction kernel per number of threads, although a breakdown of calculation costs shows a room for further improvements. Water and granular dam-break tests were also performed on the PEZY-SC system to validate the SPH and DEM codes, respectively.