J4 ›› 2011, Vol. 24 ›› Issue (5): 56-62.

• 目录 • 上一篇    下一篇

基于异构模式的云计算关键技术研究

 张庆科, 杨波*, 王琳, 陈贞翔   

  1. 济南大学信息科学与工程学院,山东省网络环境智能计算技术重点实验室,山东 济南 250022
  • 收稿日期:2011-06-30 出版日期:2011-10-20 发布日期:2011-10-20
  • 通信作者: 杨波(1965-),男,博士生导师,教授,研究方向为计算机网络与智能信息处理。 E-mail:yangbo@ujn.edu.cn
  • 作者简介:张庆科(1985-),男,硕士研究生,研究方向为高性能计算,数字水泥建模。Email:miczqk@hotmail.com
  • 基金资助:

    国家 973 计划前期研究专项基金(2010CB635117);国家自然科学基金(60873089,60573065,60673130,90818001,F020804);山东省自然科学杰出青年基金(JQ200820)

esearch on heterogeneous model based key cloud computing technologies

 ZHANG Qing-Ke, YANG Bo*, WANG Lin, CHEN Zhen-Xiang   

  1. Shandong Provincial Key Laboratory of Network Based Intelligent Computing,
     School of Information Science and Engineering, University of Jinan, Jinan 250022, China
  • Received:2011-06-30 Online:2011-10-20 Published:2011-10-20

摘要:

        结合云计算中Map/Reduce分布式编程技术引入了基于CPU-GPU异构混合并行编程模式,给出了该并行编程模式的原理和实现过程。该模式通过采用CUDA多线程并行机制提高了大规模数据处理的效率。文中对比分析了云计算中两种典型的分布式存储系统GFS和HDFS,最后从宏观角度阐释了云计算虚拟化技术的三层部署架构和基本类型。

关键词: 云计算, 图形处理器(GPU), CUDA, 并行编程模型, 分布式存储, 虚拟化

Abstract:

        This paper presents a CPU-GPU heterogeneous parallel programming model based on the distributed programming technology of cloud computing, Map/Reduce. This paper also gives its principle and implementation process. It improves the efficiency of large-scale data processing with the multi-thread parallel mechanism, CUDA. We contrastively analyze two typical distributed storage systems, GFS and HDFS. We eventually present the threelayer architecture and basic type of virtualization technology.

Key words: cloud computing, GPU, CUDA, parallel programming model, distributed storage, virtualization

中图分类号: 

  • TP393