Performance Issue Identification And Diagnosis Method Of SaaS Software Runtime Based On Log

SaaS software is deployed in a cloud computing environment and provides software functionality through cloud services.With the development of cloud computing,SaaS software has become more and more widely used.However,SaaS software will suffer performance issues at runtime due to the affect of factors such as internal structural design or external dynamic complex environment.How to timely and accurately identify and diagnose the performance issue when it occurs to maintain software performance is facing major challenges.Running log is an important means to record the current system performance parameters,which provides the important basis for identifying and diagnosing system performance issues.However,SaaS software runtime will generate a large number of logs,which contain less information about software performance.How to generate performance logs to support SaaS software performance issue identification and diagnosis is also facing major challenges.In response to the above challenges,this paper proposes a method for identifying and diagnosing performance issues of SaaS software based on running log analysis.The method can generate sufficient and necessary performance logs,and timely and accurately identify and diagnose performance issues at SaaS software runtime based on the generated logs in an automated manner.The main research work includes:(1)Design a performance log generation framework supporting the performance issue identification and diagnosis of SaaS software.The framework provides different collection services for three layers of cloud computing platform: resource layer,platform layer and application layer.The collection services acquire the performance logs of SaaS software system by means of log capture and system monitoring.Three layers of different performance metrics are analyzed to determine the performance log properties that support the performance issue identification and diagnosis of SaaS software,which can directly and comprehensively reflect the performance status of software runtime.The performance issue is defined based on the performance metric of the application layer to prepare for the labeling of the performance issue type.(2)Present a method for identifying performance issues of SaaS software based on Hidden Markov Random Field(HMRF).This method uses HMRF to model the performance state log(observations)of the system at runtime,and uses the estimation principle of Maximum a Posteriori(MAP)to solve the identification problem.Hopfield Neural Network(HNN)was used to estimate the MAP,while the Expectation Maximum(EM)algorithm was used to estimate the parameters of the HMRF model.Taking advantage of these three methods,the performance issue identification model can be established quickly and the current performance state of the system can be calculated.(3)Present a method for diagnosing performance issues of SaaS software based on Restricted Boltzmann Machine(RBM).In this method,the features of all dimensions are normalized to eliminate the impact of dimensionality,and the features of each dimension are made independent through Independent Component Analysis(ICA).Then,a SaaS software performance issue diagnosis model is established based on RBM,which is simplified into a Maximum Likelihood Estimation(MLE).The performance issue diagnosis model established by this process has higher diagnostic ability and can identify the categories of performance issues with finer granularity.(4)Conduct case study and experimental study for the method presented in this paper.In combination with a real product system deployed on the cloud computing platform,the method proposed in this paper is proved to be efficient and accurate in identifying and diagnosing the performance issues of SaaS software system at runtime.The superiority of the method proposed in this paper and the feasibility of the method in practical application are proved by comparative experiments.