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 (s1): 115-119.doi: 10.13190/j.jbupt.2014.s1.022

• 研究报告 • 上一篇    下一篇

云计算环境下MapReduce的资源建模与性能预测

乔媛媛, 刘芳, 凌艳, 尹劲松   

  1. 北京邮电大学 信息与通信工程学院, 北京 100876
  • 收稿日期:2014-01-01 出版日期:2014-06-28 发布日期:2014-06-28
  • 作者简介:乔媛媛(1987- ),女,博士生,E-mail:qyybupt@126.com;刘 芳(1968- ),女,副教授,硕士生导师.
  • 基金资助:

    国家自然科学基金项目(61072061);国家科技重大专项项目(2012ZX03002008);高等学校学科创新引智计划项目(B08004);中央高校基本科研业务费专项资金项目(2011RC0117)

Resource Modeling and Performance Prediction of MapReduce in Cloud Computing Environment

QIAO Yuan-yuan, LIU Fang, LING Yan, YIN Jin-song   

  1. School of Information and Communication Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China
  • Received:2014-01-01 Online:2014-06-28 Published:2014-06-28
  • Supported by:
     

摘要:

为了预测云计算环境下的作业资源与时间消耗,根据MapReduce的资源消耗模式,量化了MapReduce作业的资源使用,提出了一种预估Hadoop的MapReduce作业的中央处理器(CPU)利用率和运行时间的模型. 使用多项式回归的方法,可以在云计算环境下,对不同配置的MapReduce作业的CPU利用率和运行时间作出预判. 使用不同配置条件下CPU密集型的Hadoop基准测试验证了该模型的有效性,最后使用误差平方和、平均绝对百分误差、标准差和确定系数4种评估方法计算了模型预测的精准度.

关键词: 云计算, MapReduce, 资源建模, MapReduce作业耗时建模, 性能预测

Abstract:

The prediction of job running time and computing resource is very important in production environment. However, the performance of cloud computing is not easy to be predicted due to complicated computing resources in environment of cloud platform. A model is proposed based on resource consumption patterns for predicting the running time and central processing unit (CPU) resources consuming of MapReduce job in cloud computing environment. This model comes from polynomial regression modeling to predict the performance of MapReduce job. A variety of criteria is used to evaluate the model. Experiment shows that this model could predict the running time and CPU resources consuming of MapReduce job with high accuracy.

Key words: cloud computing, MapReduce, resource modeling, MapReduce job time modeling, performance prediction

中图分类号: