并行计算时间模型研究

Research on Time Models ofParallel Computing

本文系统地总结和探讨了共享和分布式存储环境下的并行计算时间模型。微观上，结合并行机结构特征和通信机制，揭示了延长算法运行时间的关键因素，并据此提出一些优化原则和效率评价准则，能辅助用户修改并行算法达到最优性能；宏观上，给出了基本消息传递的常用通信原语类型和部分原语操作时间经验公式，能辅助用户选择最优通信原语和问题粒度，正确预测程序的运行时间和性能。

Parallel computing time models are systematically surveyed and discussed in this paper.Microscopically,based on the architecture characteristics and communication mechanisms of target machines,we have discovered the key factors to significantly delay the running time of parallel algorithms,and posed some principles to optimize the algorithms,and presented a set of efficiency evaluation rules to help the user improve their algorithms to attain optimal performences.Macroscopically,based on messagepassing,we have analyzed the usual communication functions,and given out a part of experimental time formulas,which help the user select optimal communication functions and granularities,and predict the running time and performance of algorithms.。