On parallelization and load balancing aspects of the finite-pointset method
On parallelization and load balancing aspects of the finite-pointset method
复制标题
关于有限点集方法的并行化和负载平衡方面
DOI：
10.1080/00207160903452210
复制
发表时间：
2011
影响因子：
1.8
通讯作者：
Anshul Saxena
中科院分区：
文献类型：
--
作者：
A. Subramanyam;J. Kuhnert;S. Tiwari;S. Sundar;Amol Joshi;Anshul Saxena
点集法（FPM）是一种无网格拉格朗日方法。在这篇文章中，我们将讨论在分布式内存架构上并行化FPM的问题。消息传递接口（MPI）标准库用于通信。我们已经开发了算法，以实现细粒度的并行。与负载平衡和处理器之间的通信有关的问题已经得到解决。实现了BiCGSTAB线性求解器的并行版本。并行线性求解器，结合我们的负载平衡技术的效率和加速比的结果。
The Finite-pointset method (FPM) is a mesh-free Lagrangian method. In this article, we discuss issues in parallelizing the FPM on distributed memory architecture. The Message Passing Interface (MPI) standard library is used for communication. We have developed algorithms to achieve fine-grained parallelism. Issues relating to Load Balancing and communication between processors have been addressed. A parallel version of the BiCGSTAB linear solver is implemented. The efficiency and speedup results for the parallel linear solver, used in conjunction with our load-balancing technique, are presented.