A Study On The Utility-based Dynamic Resource Allocation Policy In Virtualized Environment

Along with the rapid development of virtualization technology and computer hardware platform architecture, management of computing resource has drawn many institutions and experts'attention in the research field of computer architecture in recent years. Virtualization technology can reduce the number of physic servers in the data center or large scale resource pool by server consolidation to reduce power consumption, increase server reliability and so on. It also brings challenges to the optimization of resource allocation policy due to the changes in the underlying architecture and the continuing changes of the requirement of application in the virtual machines. As the demand for resource of application is difficult to predict, the system itself is unable to monitor the actual resource demand of each virtual machine, how to allocate resource effectively and adaptively in the resource pool to improve the utilization of the computing resource and to meet the performance requirement of the application is an important research subject in the resource allocation of virtualized computing environment.This dissertation studies the existing resource allocation strategies of the resource pool, elaborates on the status of virtualization technology, Xen virtualization technology, and the related methods of resource management, and summarizes the existing resource allocation strategies in the virtualized computing environment. On this basis, two kinds of utility-based dynamic resource allocation policies in the Xen-based virtualization platform are proposed, as well as the performance mapping is established to obtain a better prediction of the application about the resource demand in the dynamic control process of the resource allocation. The main contents include:(1) In the case of the virtual machine monitor could not adjust resource allocation dynamically among virtual machines according to the load characteristics, a resource mapping model is established by the use of neural network after the collection of the massive sample data from experiments to predict the performance achievement accurately in certain resource allocation.(2) For the resource requesting party, an economic computing-based dynamic resource allocation policy is proposed based on the maximization of overall Guest Domains'utility. The price which is influenced directly by the resource utilization and the global utility function are presented to dynamically adjust the resource allocation to address the resource fierce competing from the budget-constrained Guest Domain. In the policy, the tradeoff between performance and budget is considered in the utility function of each domain, and the global utility function of the overall domains takes each domain's utility into consideration to obtain the optimal resource allocation. Furthermore, a"credit mechanism"is proposed to stimulate reasonable resource utilization and the fairness of the policy.(3) From the perspective of resource providers, each domain is classified into different service levels based on SLA service differentiation. A service differentiation based dynamic resource allocation policy is proposed. To maximize the system revenue, according to the different rate of profit and penalty when the performance objectives for each domain are met, establish the relationship between resource allocation and the system revenue, and then a non-linear continuous optimization model is obtained to meet the different requirements of each domain and improve resource utilization.In this dissertation, the simulation is executed to verify the efficiency and rationality of the proposed resource allocation policies. The results show that the economic computing-based dynamic resource allocation policy can allocate resource reasonably among the domains according to the tradeoff made by them whenever the conditions of load and performance requirements are changed. The service differentiation based dynamic resource allocation policy can provide different services to the different domains according to the load conditions, and ensure that the resource requirements of all domains are satisfied when the resource are enough. Two kinds of resource allocation policies can reasonably allocate resource dynamically, and improve resource utilization.