Recession-based The Multifractal Software Performance Prediction Methods

A number of recent measurements and studies of software system, which had run for along time continuously, demonstrated that one of the reasons of "software aging",characterized by progressive performance degradation and a sudden crash of softwaresystem is exhaustion of operation system, and they also reported that the data of resourceusage collected displayed fractal characteristics.This paper discusses measurement-based software rejuvenation strategy, combinedwith a multifractal-based method to analyze the fluctuation of the parameters of systemperformance which are monitored and collected periodically. In conclusion, it predictsresource consumption and the trend of software aging.Firstly, this study uses fractal theory to discuss the fractal structure of the parametersof system resources that influenced the performance of software system. And this thesisforecasts the fluctuation of parameter series with multifractal spectra and its parameters.And the results show that the variations of the parameters are not a stochastic process, buthave characteristic properties of fractal. In addition, the characteristics of the spectra canbe used to analyze the changes of system parameters during the running time qualitatively.Secondly, this paper uses the performance parameters data collected from a realisticsoftware system to analyze the H(o|¨)lder exponent of memory resources. The experiment hasbeen taken to calculate the multidimensional H(o|¨)lder exponent of parameter series relatedwith several memory resources. The research shows that the distribution of the exponentcontains information of many impacts of software performance, and has certain forecastingcapability, so that it can be used to realize software rejuvenation comprehensively.Finally, the Auto-Regressive Moving-Average (ARMA) model, which is the classicalmodel in the field of statistical forecast, is adopted in order to carry out the analysis of themultidimensional exponent and build the corresponding forecast model. The results ofexperiments indicate effective ability to predict software aging, and it provides a basis tohave software rejuvenation to come true.