基于MPI的矩陣乘法效率研究
[Abstract]:Parallel computing is the main theme of engineering design and scientific computing in the 21st century. It is considered as the basic method to solve problems in various fields. With the increasing of network speed, parallel computing has been transformed from experiment to computation. The typical representative of parallel computing is grid computing based on Internet and cluster computing based on LAN. Because of its excellent expansibility and cost performance, cluster computing has become the focus and mainstream of efficient computing systems. Although the cluster constructed by the parallel programming environment (MPI (MessagePassingInternet) has high computing performance, there are still many problems to be solved in the implementation of matrix multiplication in the parallel cluster based on the parallel programming environment. This paper studies and discusses the following aspects: firstly, the basic theory of parallel programming is introduced, the factors that affect the performance of the program are analyzed, and the feasible methods to improve the parallel performance in LAN are put forward. Secondly, because there are no primitives specifically for matrix transmission, the cost of matrix transmission is very high. Based on the above problem, a matrix transmission protocol is proposed to reduce the communication overhead. Thirdly, a new task scheduling algorithm is proposed in order to reduce the time consumption of parallel computing and realize the load balance of system computing nodes. Finally, the experimental platform based on LAN is constructed to test the above theoretical research, and the test results are analyzed.
【學(xué)位授予單位】:成都理工大學(xué)
【學(xué)位級(jí)別】:碩士
【學(xué)位授予年份】:2013
【分類號(hào)】:TP338.6
【參考文獻(xiàn)】
相關(guān)期刊論文 前10條
1 黃淑玲;可擴(kuò)展并行計(jì)算的應(yīng)用與研究[J];電腦知識(shí)與技術(shù);2005年12期
2 石瑩瑩;李濤;;基于MPI的Strassen矩陣乘法算法的并行計(jì)算研究與實(shí)現(xiàn)[J];電腦知識(shí)與技術(shù)(學(xué)術(shù)交流);2007年22期
3 楊愛(ài)民,陳一鳴;MPI并行編程環(huán)境及程序設(shè)計(jì)[J];河北理工學(xué)院學(xué)報(bào);2005年03期
4 孫振河,李金寶,任美睿;網(wǎng)格計(jì)算環(huán)境下基于任務(wù)池的任務(wù)調(diào)度方法[J];黑龍江大學(xué)自然科學(xué)學(xué)報(bào);2005年01期
5 徐萬(wàn)鴻,宋佳興,劉衛(wèi)東,徐恪;基于節(jié)點(diǎn)機(jī)計(jì)算能力的網(wǎng)絡(luò)計(jì)算體系[J];計(jì)算機(jī)工程與應(yīng)用;2001年16期
6 曾麗芳,曾勁松;DSM體系結(jié)構(gòu)對(duì)并行編譯系統(tǒng)的支持與挑戰(zhàn)[J];計(jì)算機(jī)工程與應(yīng)用;2003年01期
7 趙晨,李仕鋒,許小剛,王萃寒;網(wǎng)絡(luò)環(huán)境中MPI和PVM的分析與比較[J];計(jì)算機(jī)工程與應(yīng)用;2003年03期
8 鄭磊,周濟(jì)華;單邊通信在標(biāo)準(zhǔn)Unix系統(tǒng)上的實(shí)現(xiàn)[J];計(jì)算機(jī)工程與應(yīng)用;2004年16期
9 馬艷琨,馬勝甫,田俊峰,張建勛;一種用于PC存儲(chǔ)集群的動(dòng)態(tài)負(fù)載平衡策略[J];計(jì)算機(jī)工程與應(yīng)用;2004年29期
10 周恩強(qiáng),趙軍鎖,楊學(xué)軍;MPI及MPI的高效實(shí)現(xiàn)[J];計(jì)算機(jī)工程與科學(xué);1999年05期
相關(guān)博士學(xué)位論文 前1條
1 蔣江;異構(gòu)集群系統(tǒng)中基于進(jìn)程遷移機(jī)制的負(fù)載平衡算法的研究[D];國(guó)防科學(xué)技術(shù)大學(xué);2002年
相關(guān)碩士學(xué)位論文 前7條
1 鄒暉;異構(gòu)機(jī)群并行計(jì)算環(huán)境研究及應(yīng)用[D];成都理工大學(xué);2002年
2 劉輝;基于MPI的并行程序設(shè)計(jì)技術(shù)[D];鄭州大學(xué);2002年
3 李小衛(wèi);并行程序動(dòng)態(tài)行為可視化監(jiān)測(cè)及并行I/O研究[D];成都理工大學(xué);2003年
4 王春霞;PC集群系統(tǒng)的并行開(kāi)發(fā)環(huán)境MPI及其程序性能的分析研究[D];鄭州大學(xué);2003年
5 韋祥文;MPI平臺(tái)下二維歐拉方程數(shù)值解法[D];西北工業(yè)大學(xué);2003年
6 馮憲澄;基于MPI的并行小波圖像分割研究[D];浙江工業(yè)大學(xué);2004年
7 汪梅婷;基于MPI的并行計(jì)算中矩陣傳輸協(xié)議及負(fù)載平衡算法的研究與設(shè)計(jì)[D];燕山大學(xué);2006年
本文編號(hào):2314774
本文鏈接:http://sikaile.net/kejilunwen/jisuanjikexuelunwen/2314774.html