Research On Reliability Evaluation Model And Prediction Method Of Dynamic Combination Cloud Service

With the development of the Internet and the Cloud computing,service-oriented computing is standing out of the current computing modes.Cloud computing is one of the SOCs,which focus on the physical hardware resources to the resources pool uses the virtualization technology,and provides different kinds of resources services to the different users.Whether the cloud services can provide reliable calculation and application resources services is the important key to the survival and long-development of the cloud services.With the continuous development and demand of service requesters business growing,the cloud service system needs to combination some single function service to a combination service which has some additional functions to need the users requirements,and the services must has a high reliability.Cloud computing has some characteristics such as large size,distributed,loose coupling and the complex network environment.The most important problem that the cloud computing faced is how to ensure the combination services has high reliability after combination.The reliability is one key value to assessment the non-functional attributes of services.The high and low of the services reliability directly determines that whether the services can complete the users'business requirements in a specific environment.To some extents,increasing the reliability of services also can decrease the service provides'costing.The reason of the failure of cloud services is manifold because of the large size of the cloud computing system.And the traditional reliability evaluation and prediction methods can't apply to the now cloud service system.In this paper,we analyze the characteristics of the cloud services and the problems the cloud services faced and have a summary of the present research condition of the software and hardware reliability.Have a deep analyze and research of the service reliability in terms of the performance indicators,failure characteristics,mode building,assessment prediction,feedback update and give a corresponding mode a prediction algorithm.In the end,we do some experiments to validation the assessment mode and prediction algorithm we put forward.The main work of this paper is divided into the following four aspects:?1?Have an analysis of the service-oriented computing and reliability and have a conclusion of the current model building of reliability and assessment theory.Put forward the problem which need to resolve and the main defects of the services'reliability and prediction.And these analyzes lay a foundation for the following theoretical research.?2?For reliability calculation theory when calculating the dynamic and complex and changeful cloud service reliability,environmental adaptability problems,combining the theory of stochastic Petri net theory and CCSPNet model transformation and put forward the model about the reliability of cloud service with the service binding graph.Based on a deep considering about the role of the service requester in the service interaction model put forward a prediction model of reliability of the cloud services based on the credit awareness in both sides in the model.Fully consider the both sides of the service interaction model,through computing the satisfaction of the users and services and put forward a service reliability forecast update model based on the credit awareness.Have an analysis of the service combination's model building based on the Petri theory and CCSPNet theory,and give the computing method of the service reliability.Considering of the dynamic binding and update in the process of the service combination,give the concept of service binding graph and the computing method of service reliability.Based on the geographical position information of users and the function operation description information of service,give the computing method of users'similar sets and services'similar sets,and the method of credibility between similar users or similar services.At last this paper put forward a service reliability assessment and prediction model based on the credit awareness.?3?Aiming at solving the high algorithm complexity and the high cost of historical data,put forward a prediction algorithm based on the improved Bayesian algorithm—IDLM.Give the indicator system of the service reliability through analyzing the characteristics of the service combination.Have a deep analysis of the problem of the traditional Bayesian algorithm,and use the index of weighted regression method to resolve the state error variance in the process of the algorithm.This method also increases the precision of the prediction algorithm.In combination with the prediction and update model which is gave up in the 3^rd chapter,update and optimize the prediction results through computing the satisfaction of the both sides of the users and services.?4?Test the algorithm and model that we put forward with the cloudsim,and analyze the experiment result with matlab.Do the verification experiment to test the prediction performance about the prediction model and the prediction algorithm IDLM.The results of the experiment show that:the prediction model has a good performance in prediction of reliability,and at the same time we introduce the reputation value of the both sides and provide a reference for the next prediction;the other is that the IDLM algorithm has a higher prediction accuracy and has low consumption of history data.