软件老化的多元时间序列分析方法  

Multivariate Time Series Analysis of Software Aging

作　　者：郑鹏飞[1] 齐勇[1] 陈鹏飞[1] 

机构地区：[1]西安交通大学电子与信息工程学院计算机软件与理论研究所,西安710049

出　　处：《计算机科学与探索》2012年第2期125-133,共9页Journal of Frontiers of Computer Science and Technology

基　　金：国家自然科学基金No.60933003~~

摘　　要：针对目前软件老化分析中的单参数模型,以及未考虑变量间关联性和影响性的多参数模型的不足,提出了运用多元时间序列模型分析软件老化的方法。通过对实验采集的HelixServer-VOD服务器性能数据的分析,运用格兰杰因果性检验,证实了软件老化发生和发展过程中各个性能参数间存在显著的相互影响性。引入向量自回归模型对软件老化进行建模,给出了软件老化在多个参数维度的联合预测以及参数间相互影响方式的定量描述。通过模型的迭代计算,比较了向量自回归模型与现行的未考虑参数间相互影响的模型对多个性能参数变化曲线的拟合及预测情况,证实了VAR模型更接近软件老化的本质。Currently, models adopted to analyze software aging are either univariate models or multivariate models which do not incorporate mutual influence or correlation of the model parameters. To extend these models, this paper introduces a multivariate time series approach to analyze software aging based on Granger causality test and vector auto-regression （VAR） model. Firstly, through the experiment of the HelixServer-VOD system, Granger causality test is used to verify that there exists mutual influence between performance parameters during software aging. Secondly, VAR model is constructed to depict and forecast software aging progress in multiple parameter dimensions. Then by comparing VAR model to prevailing software aging models in an empirical way, the results indicate that VAR model gets nearer to the nature of software aging than others.

关 键 词：软件老化 向量自回归 格兰杰因果性检验 HelixServer 时间序列分析 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]