Analysis Of Performance And Power Consumption For Error-prone Elastic Clouds

Cloud computing is a kind of computer technology.At the same time,it is also a new type of business model.It sells computing power to users in the form of commodities,and charges on demand.It is mainly has three service modes.From the bottom up,Infrastructure as a Service,Platform as a Service,and Software as a Service are provided.As the lowest-level infrastructure-as-a-service,its quality of service directly determines the quality of the above two layers of service.Therefore,the analysis of the performance and power consumption of the IaaS cloud system has become one of the research hotspots of the current cloud computing.Due to the complexity of cloud computing,analyzing the performance and power of the cloud system is a challenging and difficult research task.In particular,many current cloud computing systems support elastic scaling,which increases the difficulty of analysis and reduces the accuracy of analysis.Currently,there are two analysis methods for performance and power consumption analysis of cloud computing systems,which are measurement-based and model-based analysis methods.Integrating various factors,this paper uses a modeling approach to evaluate the performance and power consumption of the elastic cloud system.Based on other cloud computing performance and power analysis studies,this paper presents a comprehensive analysis method to analyze the performance and power consumption of the elastic cloud.First,by analyzing the service request and task processing flow of the cloud system,an abstract model of cloud system task processing is established.The elastic scaling of the system,the virtual machine error,physical machine error and task retry are considered and the model is more in line with reality cloud computing system features.Then employ a tedious state-based Markovian model to obtain exact solution of performance and power consumption metrics.This article defines the following indicators to evaluate system performance and power consumption: expected instantiation response times,power consumption rate,task rejection rate under different load conditions.To verify the proposed method,we conducted experiments in a campus cloud platform environment and compared it with traditional performance analysis methods.The 90% confidence interval of the above analysis indicators was calculated by collecting log data,and compared with the theoretical values of the model and the comparison model,which proved the reliability and accuracy of the model.Finally,based on the deduced theoretical model,the cloud systems which different number of physical machines are observed respectively by controlling the changes in the arrival rate,error rate,and elastic scaling rate of the tasks,and the trend of the above three assessment indicators is changed.This is the theoretical guidance for the selection of the number of physical machines for different parameters in the actual deployment of the cloud system.