北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2014, Vol. 37 ›› Issue (5): 20-25.doi: 10.13190/j.jbupt.2014.05.005

• 论文 • 上一篇    下一篇

基于CMM模型的HDFS负载均衡策略

卢美莲1, 朱亮亮2   

  1. 1. 北京邮电大学 网络与交换技术国家重点实验室, 北京 100876;
    2. 华夏银行股份有限公司 科技开发中心, 北京 100005
  • 收稿日期:2013-12-18 出版日期:2014-10-28 发布日期:2014-10-28
  • 作者简介:卢美莲(1967- ), 女, 副教授, E-mail: mllu@bupt.edu.cn.
  • 基金资助:

    国家科技重大专项项目(2012ZX03005010-003);国家高技术研究发展计划项目(2014AA01A706)

Load Balancing Strategy Based on CMM Model in HDFS

LU Mei-lian1, ZHU Liang-liang2   

  1. 1. State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing 100876, China;
    2. Science and Technology Development Center, Hua Xia Bank Limited Company, Beijing 100005, China
  • Received:2013-12-18 Online:2014-10-28 Published:2014-10-28

摘要:

为实现Hadoop分布式文件系统的负载均衡,并保证较低的负载迁移代价和数据传输代价,提出了确定环境下多阶段多目标(CMM)决策模型. 该模型以CPU、内存和磁盘剩余负载能力作为决策条件,以负载均衡效果、负载迁移代价和数据传输代价作为决策目标,依据决策节点间的影响关系构建有向无环图,通过多个决策阶段的决策,并计算方案效用确定最优均衡方案. 仿真实验结果表明,基于CMM模型的负载均衡策略能取得较好的负载均衡效果、负载迁移代价和数据传输代价.

关键词: 分布式文件系统, 负载均衡, 决策模型, 剩余负载能力

Abstract:

In order to balance the load of Hadoop distributed file system with lower load migrating cost and data transmission cost, a certainty multi-stage and multi-object (CMM) decision model was proposed. The model is a directed acyclic graph built on decision nodes, which adopts the remaining load capacities of CPU, memory and disk as decision preconditions, and also adopts load balancing effect, load migrating cost and data transmission cost as decision targets. By CMM model, the best balancing plan is determined by selecting results of multiple decision stages and computing the plan usage. Simulations show that the CMM based strategy can achieve better load balancing effect, load migrating cost and data transmission cost.

Key words: distributed file system, load balancing, decision model, remaining load capacity

中图分类号: