Research On Cloud Service SLA Compliance Verification And Performance Optimization

As next generation of computing model and Internet applied model, cloud computing provides high reliability, high availability, high scalability and flexible customized service in a cost effective way. In recent years, cloud computing has gained great development, and there are growing concerns about the cloud computing in academia and industry with advances in technologies and ever-growing service demand.However, virutalization technologies used in cloud computing widly not only provide dynamic scalability of resource, but also separate ownership and use right of the resource in cloud. Clients can not directly control physical resource. Hence, cloud users can not know the cloud resource properties. This has led to security problem that the cloud resource provided by Cloud Servie Provider(CSP) may does not meet the requirements of Service Level Agreement(SLA), and this problems will threaten mutual trust between cloud user and CSP. It is doubtful whether the current cloud service is compliance with SLA for cloud user. This discourages the use of cloud services. On the other hand, high energy-consuming is one of the typical features of cloud computing. The CSPs want to decrease operating cost and get more profits urgently by effectiveness optimization method under service level agreement. At present, there are still some problems in SLA evaluation and effectiveness optimization which should be solved more urgently. This dissertation focuses on the service level evaluation and effectiveness optimization on cloud. The main work and contributions in this dissertation include:First, CSP can assign physical memory to virtual machine dynamically through the memory management mechanism of Xen. So, the physical memory assigned to virtual machine owns better dynamic performance. Untrusted CSP may assign physical memory by reduced service level approach in the pursuit of greater economic profits. Hence, a light-weight hierarchical memory verification mechanism in virtual machine running environment is proposed. The service capacity SLA of physical memory assigned to guest virtual machine is audited within a certain precision through current memory usage and phycial memory rapid measurement strategy, without introducing heavy overhead to guest virtual machine.Second, we propose a statement-level resource consumption model of user application. First, the historical observation sample space is obtained throuth replaying the user application on local platform by mass small dataset input. Then, the resource comsumption model between input dataset of user application on local and resource consumption using regression approach. Finally, we get the performance difference on service capacity between local platform and cloud platform by benchmark application. The worst-case resource consumption can be estimated relay on basic block contral flow graph. And the cloud service capacity SLA can be verified by comparing the actual resource consumption of user application with its estimated worst-case resource consumption.Third, we define and extend the cloud service capacity representation strategy, that is, the mapping rules between cloud service capacity and SLA specification were given. Then, we propose the service capacity SLA evaluation scheme based on the cloud running time environment service capacity captured by cloud resource service collecting unit in running time environment and the resource utilization by cloud user. Finally, we design the cost optimal service measurement strategy by economic indicator and performance overhead brought by service measurement process. Meanwhile, the frequence of service capacity measurement can be readjust dynamically by the feedback of cost optimal service measurement unit.Fourth, it makes a comprehensive analysis of power consumption, performan ce, reliability in cloud computing environment from the view of virtual machine power consumption, performance, software and hardware failures, and failure recovery perspectives. First, the correlated model between virtual machine performance an d power consumption is build under service reliability SLA constraint. Then, the power consumption and the task completion time about the given task were balanced. And the optimal processor frequency is proposed. Finally, the availability of the correlated model on balancing the relationship among power consumption, perfromance and realiability is verified through some experiments.