Research On The Adaptive Behaviors Prejudgment Method Of The Cloud Service System Performance Self-Optimization

In the cloud computing era,with the increasing sophistication of software systems,the optimization ability of the software system that running in the open dynamic and unstable environments is more and more important.Because of the new characteristics of the cloud computing and the new business model which is called pay-on-demand,the cloud service system is not only required to have the ability to meet the application requirements with the minimum cost,but also required to have the ability to self-adjust when its actions are far from the expected.The adaptive method of the service performance in the cloud environment have been one hotspot of the service computing and cloud applications research field.But in the Sense-Decisionmaking-Implementation,the traditional cloud service performance self-adapting research framework,the trigger-events,which lack the ability of self-evolution,are artificial designed based on the SLA that could not to cope with the complex cloud environment now.On the other hand,the traditional election process of the best performance self-adapting action type should work out the action sequences firstly and then implement the least-costs and highest-benefits action sequence by the valuation model of costs and benefits that waste much resource.In this thesis a new process procedure,Prejudgment,is added to the traditional research framework,and a new research framework,Sense-Prejudgment-Decisionmaking-Implementation,is proposed.The new process contain two parts,one is the judge process of trigger-event,and another is the selection process of the best performance self-adapting action type.In order to solve the problems of the trigger-events in the traditional research,the thesis proposed the new concept of the warning-event based on the traditional trigger-event and completed the design of the warning-event generating algorithm based the component history data.In order to reduce the resources consumption and improve work efficiency,the thesis established the selection model of the best performance self-adapting action type by the deep learning algorithm that can choose out the best performance self-adapting action type before the action sequence generated.The effectiveness of the warning-eventgenerating algorithm and the best performance self-adapting action type selection model training algorithm are validated by the experiments.The experimental results show that,the warning-event proposed in the thesis have the ability of self-evolution and they are triggered before the service performance falling to the SLA limits.And the experimental results also show that,the selection model of the best performance self-adapting action type can choose out the best performance self-adapting action type by the real-time environmental information before the action sequence generated.The accuracy and effectiveness of the method presented in the thesis are validated by series of experiments.