摘要
文中工作是在曙光２０００ 上的ＰＶＭ 环境中为自动并行识别系统（ＡｕｔｏＰａｒ）实现的，介绍的是数据收集函数Ｃｏｌｌｅｃｔ算法及其设计优化以及对原有算法所作的结构性改进，主要思想是将存储转发机制引入到数据收集算法中，从而省去了大量的数据填充及抽取操作，并防止了由此产生的数据收集错误．众所周知，通信问题在并行机系统中是关键问题，因此，Ｃｏｌｌｅｃｔ效率的高低直接关系到ＡｕｔｏＰａｒ的加速比．采用本算法后。
The work in PVM's environment on the Dawning\|2000 system is realized for the automatic parallel recognition system(AutoPar). Introduced here is the data collection function Collect's algorithm and its design optimization and the structure improvement of the original algorithm.Collect algorithm and its design in AutoPar's parallel communication library based on PVM are studied,and a new algorithm and store and forward mechanism are found to implement Collect,which is faster than its original one.Because it emphasizes particularly the communication among processors, Collect will influence the speedup of AutoPar at Dawning\|2000. Its major idea is to introduce the save\|transmit mechanism into the data collection algorithm, reducing a large number of data fills\|in and draws\|out operation and preventing the data collection mistakes from being produced. As everyone knows, the communication problem in running parallel machine system is a vital problem, so the height of Collect's efficiency directly concerns how the quickening of AutoPar is compared. After the algorithm is used, its efficiency is doubled through real testing.
出处
《计算机研究与发展》
EI
CSCD
北大核心
1999年第10期1202-1206,共5页
Journal of Computer Research and Development
基金
国家高技术"八六三"项目基金
关键词
存储转发
编译程序
自动并行库
通信
优化
parallelization, AutoPar, PVM, collect algorithm, store and forward
