A Method To Evaluate Web Application Reliability Based On Workload

Because of the complex Web application environment, systemaccess have periodical change according to the time interval, thereasons result in Web page failure are also various, not only the softwareinternal failures, but also the user usage and network environment. Sothe traditional measuring methods of software reliability are facingmany new challenges.Generally the traditional software reliability can be evaluatedquantitatively by using reliability model, but most of the models arebased on software internal failures, ignoring the user interactivebehaviors with the system. Therefore, the traditional method based onfailure data to measure the Web application system reliability can notbe good to meet the demand. In this paper, we give a new methodbased on Web application workloads to improve the accuracy usingthe traditional software reliability models for the Web application systemreliability measurement.For a Web application system, log files usually stored on Web server.These log files record the interactive behaviors between users andsystem in detail. Firstly, this paper analysis the log format of the IIS andApache server which are commonly used today. We get the Webapplication workload information and error information from these logfiles, through the log data preprocessing. This paper use specificworkload information includes hit count, user count, session count andbyte count. This four workload measures are the quantitative indexs ofinteractive behaviors about hit, form sumbit, session and access flow;The error information includes error count, error hit count, error usercount, error session count and error byte count produced in theinteractive behaviors. We use the workload to replace the timeparameters in the formual to measure the current reliability of the Web application system, according to several important indicators of thesoftware reliability measurement. Then, this paper select three classicalreliability models, they are Nelson model, J-M model and G-O model.We use the workload to replace the time and count parameters in theformual to predict the reliability of the Web application system.By the experiment results, it is demonstrated that the method basedon workload measures proposed in this paper can be more stable toevaluate the reliability of Web application system. And through threemodels, we use the traditional failure measures and new workloadmeasures to estimate parameters, then, we use them to predict thesystem’s future performance. The results demonstrated the accuracy ofour new method is better by the comparison of fitting curves betweenestimate values and actual values. At the same time, we also found theassessment results of four workload measures is different in thisexperiment. We need to do reasonable choice by the specificenvironment of the Web application.