摘要
针对网格划分的并行思想,以大型矩阵乘法为例,讨论了机群下并行子任务间分配策略的优化.描述了对矩阵实行网格划分的基本方法和并行算法的设计过程,并研究了在机群环境下的实现.分析了非理想状态下原有算法存在的不足,主要讨论了在网格阵列处理机上,如何对并行子任务进行优化分配,提出了一种合理的分配策略.通过实验结果分析,验证了该思想的正确性和实用性.
In view of the paralle thought of grid division, take large-scale matrix multiplication as example, discussed the optimization of parallel sub-duties assignment in the culster. Described the method of matrix grid division and the practice of parallel algorithm, and examined the circumstances of its formation in the cluster. Analyzed the insufficiency that the original algorithm existed, mainly discussed on the grid array processores, how to optimized assignment paralld sub-duties, proposed a kind of reasonable assignment strategy. Through the analysis of experiment results, confirmed its Correctness and usability.
出处
《微电子学与计算机》
CSCD
北大核心
2008年第3期134-136,140,共4页
Microelectronics & Computer
基金
辽宁省教育厅科研基金项目(05L209)
关键词
矩阵乘
并行
网格划分
任务分配
MPI
matrix multiply
parallel
grid division
duty assignment
MPI