摘要
研究确定了网格单元邻接矩阵带宽与并行计算效率间的关系,得到了可以通过减小单元邻接矩阵的带宽,以减小外部通信量,从而达到提高并行计算效率的目的。为了提高并行计算效率,基于Akhras-Dhatt(AD)算法思想,通过引进类似节点商的单元商对单元编号进行优化,使得矩阵带宽减小,从而设计了一种减小单元邻接矩阵带宽的AD算法。进一步基于减小矩阵带宽的AD算法,提出一种区域剖分算法。利用此区域剖分算法,对全球海域无结构网格中的部分区域进行区域剖分研究,结果表明该算法能得到很好的加速比与并行效率,计算耗时相对较短,证实利用该方法对大规模网格区域进行分裂具有可行性与有效性。
The relationship between the grid cell connectivity matrix bandwidth and the efficiency of parallel computing is confirmed, when the bandwidth is reduced, the external communication is reduced and the parallel efficiency is improved. In order to improve the parallel efficiency, based on the AD algorithm, an algorithm is developed for reducing the connectivity matrix bandwidth. This algorithm uses cell ponderation to optimize cell nodes. A domain decomposition method is developed based on this new algorithm. Parts of global ocean unstructured grid are researched based on this method, according to a series of researches, the speedup and efficiency are given. The feasibility and efficacy of the model for large-scale grid domain decomposition are shown.
出处
《江南大学学报（自然科学版）》
CAS
2015年第2期202-208,共7页
Joural of Jiangnan University (Natural Science Edition)
基金
中央高校基本科研业务费专项项目(2014B06314)
关键词
区域剖分
并行计算
AD算法
无结构网格
domain decomposition, parallel computing, AD algorithm, unstructured grid
