高可扩展格子Boltzmann方法

格子Boltzmann方法(Lattice Boltzmann Method,LBM)是计算流体力学中的一种常用方法.由于LBM中的格点仅与相邻的格点间存在数据传递,因此具有良好的并行性.LBM并行算法中的数值通信部分通常采用的是格点上的微观量——分布函数.每次传递的分布函数具有多个不同的速度方向,为了进一步减少LBM并行算法的通信开销,从格子 Boltzmann方法的物理特性以及相应的串行程序为切入点,深层次地挖掘可并行的因子,设计了专门用于通信面的类,使用数量较少的宏观量进行通信,降低了通信所占的比重,缩短了通信时间,提高了加速比和效率.实验表明,在4 096个计算核上依然有良好的加速比和效率.

Investigation of High Scalable Lattice Boltzmann Method

Lattice Boltzmann method (LBM) is a commonly used method in computational fluid dynamics.Since the lattice points in LBM only have data transfer between neighboring lattice points,it has good parallelism.The microscopic (distribution function) is the main communication data in parallel algorithm of LBM.The distribution function has a plurality of different speed directions.In order to further reduce the communication overhead of the LBM parallel algorithm,combined with the physical characteristics of lattice Boltzmann method and serial program,deep-level mined parallel factor is presented.Acommunicationclass is designed by using to communication reducing the communication time and improving the speedup and efficiency.The experimental results show that the speedup and efficiency is also better through the experiments on 4,096computing cores.