Parallel Algorithm for 3D Particle Simulation using CM-2

本文采用扫描向量模型对等离子体粒子胞内（PIC）仿真代码进行了严格的数据并行处理，提出了一些新的数据并行实现方法。自并行计算机出现以来，粒子池（PIC）粒子模拟已被公认为流体动力学和等离子体科学等学科的科学家可以用来研究空气分子或亚原子离子和电子等粒子的复杂动力学的实用工具。作为美国政府机构的“大挑战”问题之一，PIC代码的并行化已经得到了推广，因为PIC代码被认为具有固有的并行性。通常的PIC代码将空间模拟域映射到网格上。粒子被表示为在网格内移动，同时由网格点和粒子跟踪的属性都被更新。在并行计算机上，可以将粒子或网格作为主要数据结构嵌入处理器。在第一种情况下，每个处理器负责跟踪与指定粒子相关的属性，我们将其称为粒子数据。在网格点上跟踪的属性，我们称之为网格数据，可以提供给粒子，这就是所谓的计算的“收集”部分。另一方面，如果选择网格作为主要数据结构，则每个处理器负责跟踪模拟域的一个子部分，并且粒子数据可供网格点使用，这就是所谓的“分散”计算部分。在网格处理器和粒子处理器中，都需要仔细考虑负载平衡和通信成本。在我们的研究中，我们成功地实现了SPMD版本的扫描矢量模型PIC代码。

This report presents some new data parallel implementations of a plasma particle-in-cell (PIC) simulation code using a scan vector model, which is a strictly data parallel.Ever since the emergence of parallel computers, particle-in-cell (PIC) particle simulation has been recognized as a practical tool that scientists in disciplines such as fluid dynamics and plasma sciences can use to study the complex dynamics of such particles as air molecules or sub-atomic ions and electrons. Noted as one of the 'Grand Challenge' problems by U.S.Government agencies, the parallelizations of PIC codes have been promoted because PIC codes are assumed to possess inherent parallelism.A usual PIC code maps a spatial simulation domain onto a grid. Particles are represented as moving within the grid, while both the properties that are tracked by the grid points and by the particles are updated. On a parallel computer, one can embed either the particles or the grids onto the processors as the primary data structure. In the first case, each processor is responsible for tracing the properties associated with the assigned particles, which we will call the particle data. The properties tracked at the grid points, which we will call the grid data, are made available to the particles, which is so-called a 'gather' part of the computation. On the other hand, if the grid is chosen as the primary data structure, each processor is responsible for keeping track of a subsection of the simulation domain, and the particle data are made available to the grid points, which is so-called a 'scatter' part of the computation. In both the grid and particle processors, the load-balance and the communication costs should be carefully considered. We have successfully implemented a SPMD version of Scan Vector model PIC code in our research.