Research On Performance Evaluation Method For Hybrid Deployed Cloud

The widespread application of cloud computing has attracted more and more companies and organizations to deploy traditional online services to cloud platforms.At the same time,the rapid development of big data has also brought a large number of offline computing demands.In order to save costs and improve the utilization of cloud platforms,users of the cloud platform began to deploy online services and offline jobs simultaneously to a cloud platform,forming a hybrid deployed cloud environment.For the performance of the hybrid cloud platform,the service quality of the online services and the utilization of the resources of the system become the main concerns.This thesis proposes a set of reasonable and easy-to-use performance testing methods for this kind of hybird cloud scenario,and evaluates the performance of the cloud platform's hybrid part by testing the impact of resource utilization on the online service's response latency in the cloud platform.The test method proposed in this thesis includes a benchmark test assembly(Benchmark Suite).The assembly consists of four modules,which are online service deployment module,offline job deployment module,system monitoring module,and data collection module.The online service deployment module can deploy a variety of online service type applications and includes load data derived from the actual production environment.The offline job deployment module can automatically deploy common big data processing frameworks,run common big data processing programs,and include the corresponding load generation tool.The specific implementation of the two application deployment modules adopts the method of mirror integration,which facilitates the rapid environment construction and transplantation.The system monitoring module allows the user to keep track of various system resource usage during the test.The data collection module is responsible for collecting test data,which mainly includes the latency time of the online service and the various resource usage of the system.Test descriptions and load descriptions are given based on test requirements and testing can begin.After the test is completed,the collected online business latency time and system resource utilization are plotted as a scatterplot and regression analysis is performed.The obtained functional relationship can reflect the hybird performance of the cloud platform.Based on the above test methods,this thesis proposes a value computing model for the hybrid cloud platform.Substituting the tested and analyzed function of hybrid performance into the model can calculate the resource allocation that maximizes the value of the cloud platform.In addition,this thesis also proposes a virtual machine placement method to improve the performance of the whole hybrid cloud platform.This method is to improve the cloud platform by analyzing the performance difference of each physical machine in the cloud platform system and then performing load balancing.