Research On A Dynamical Component Replicas Add Method For Performance Adaptive Optimization Of Cloud Service

It is important that software system improves the self-optimization and adaptive ability with the complicated software system and opening,dynamic and uncontrollable running environment.Cloud computing has the characteristics of utility computing and pay-for-use.Therefore,the service system could not only satisfy dynamic resource allocation,but also ensure the minimum resource cost.It also could dynamically adjust resource when the self-adaptive action deviates the desired path to offer the serivices which satisfy users expectations.Traditional service system adaptive methods have some shortages in the optimization stage:(1)The research focuses on how to prevent SLA violation only based on service instance individual level.(2)The research ignores the interference which is caused by resource shared among different service instances.(3)The research ignores the cost and benefit between user demand and resource cost.Cloud service system is composed of multiple components and the component which is a basic unit deployed in the cloud virtual machines pool works together.The adaptive components form the adaptive serivice instance.The component replicas could share the application user concurrency and ensure cloud service system provides continue,stable and effective service to improve service performance.This thesis researches the component replica dynamic increase method to optimizate the cloud service performance from the view of whole performance optimization.It includes three important research aspects:prediction application user concurrency,selection increase components replicas and location the components replicas based on the theoretical research,technology research and experiment research method.First of all,research application user concurrency prediction method oriented to different concurrency characteristics by analysing autocorrelation coefficient.Then,select increase component replicas based on comproved PSO algorithm by AOE network composition algorithm.What's more,establish CC-RR model by analysing component concurrency and resource utilization.Locate component replicas based on cost and benifit model.Finally,verify component replicas dynamic increase method oriented cloud service system self-adaptive optimization validity and correctness.Experiment results show that the component replicas dynamic increase method oriented to cloud service performance adaptive optimization could accurately predict the application user concurrency and adaptive decision effectively optimization method.It could not only reduce cloud service system response time and improve service system performance,but also ensurance resource utilization satified SLA constraints,guarantee the system to provide sustainable,stable service.