Research On Management And Control Technology Of The CPU Resource In Virtualization System

Efficient use of resources has been the focus of virtualization technology，and processor isthe core resource in virtualization system. Effective management and use of the processorresources will significantly enhance the performance of the virtualization platform. With theimprovement of chip integration，the number of cores in multi-core processors is increasingwhich provides a wealth of computing resources. Besides the processor resource has got somenew features because of the use of hardware-assisted virtualization technology. The managementand use of the processor resource in virtualization system faced new challenges based on currenthardware conditions.A variety of dynamic management and control strategies were introduced to improve themethods of virtual machine resource management, and have received good effect. Even thoughthere still exist the following problems. One of them is that the multi-core computing resourcehave not been used effectively yet. Another is that the virtual machine scheduling algorithmcannot adapt to the load characteristics properly. In addition, it is difficult for user to customizethe virtual machine execution environment conveniently.This thesis proposed a multi-level processor resource dynamic configuration model, inwhich the resource granularity is divided into three categories: physical processor, VCPU and theVCPU running time. The model could meet the processor resource adjustment in demand byusing the resource dynamic adjust strategy. A new virtual machine scheduling algorithm LCVSbased on the analysis of the virtual machine load characteristics has been designed to improvethe performance of the I/O-bound virtual machines, and the algorithm guarantee the schedulingfairness at the same time. And a mechanism to support the dynamic configuration of the virtualmachine execution environment has been proposed by control the use of the processor withIntel-VT technology support, from which can make the virtual machine to have differentrun-time features. The mechanism enhanced the availability and flexibility of the VMM.The test results showed that, the dynamic configuration model can effectively improve theperformance of the virtual machine compared to the static configuration method, the settings ofadjust observation interval can make the adjustment more timely and accurately. The LCVSalgorithm can decrease the I/O-bound virtual machine response time by50%to60%and balancethe use of the physical CPU between CPU-bound virtual machines and I/O-bound virtualmachines. Compared to the traditional way, the virtual machine execution environment dynamicconfiguration mechanism can enhance the availability of the VMM.