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

Abstract: 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 dimen-sions. 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.

Key words: software aging, vector auto-regression, Granger causality test, HelixServer, time series analysis