摘要: 针对高性能计算机实际获取性能低的突出问题,在大规模科学计算和工程应用中对占主要作用的聚合通信进行了研究。通过对聚合通信Reduce算法的理论分析与试验比较,找出LogGP模型和α+nβ模型的关系表达式,并在描述聚合通信算法的理论时间复杂度时,提出了对LogGP模型和α+nβ模型的补充和改进,这些工作对聚合通信算法的理论研究和设计具有重要的意义。
关键词:
聚合通信Reduce,
LogGP模型,
α+nβ模型
Abstract: According to the most important problem of the current high performance computer that can not achieve the satisfied performance,collective communications as the most important part of the large-scale scientific computing and engineering applications are selected as the Research object.The relationship between LogGP model and α+nβ model is established through analyzing the collective communication Reduce theoretically and experimentally.The LogGP model and α+nβ model are improved when characterizing the time complexity of the collective communications.These are contributed to the studying and designing of the collective communications.
中图分类号:
黄度,刘洋,曹建文,李玉成. 聚合通信中LogGP模型和α+nβ模型的改进[J]. , 2005, 31(22): 71-.
HUANG Du,LIU Yang,CAO Jianwen,LI Yucheng. Improvement of the LogGP and α+nβ Models in Collective Communications[J]. , 2005, 31(22): 71-.