Research And Application Of Virtual Machine Live Migration Techniques Based On Xen

In recent years, virtual machine technique is applied to resources service integration, system security, distributed computing and the emerging cloud computing and other fields widely, because it can shield the heterogeneity characteristics of the underlying hardware, improve resource utilization, enhance flexibility and scalability of system, and reduce the hardware investment and management costs and so on. As an important feature of virtual machine technique, live migration of virtual machine can move running virtual machine from one physical host to another completely and seamlessly, realizing operating system and application software live migration, with the downtime in millisecond. Live migration of virtual machines is the core and key technologies of implement load balance between physical machines in platform and achieving high availability and high reliability, which is also one of hot topics in current virtualization and cloud computing related fields. On the basis of intensive analysis of Xen virtual machine architecture, we study live migration technique of virtual machine and management framework in cloud in this thesis, having done the following specific works.Firstly, we study the technique of virtual machine and live migration of Xen intensively. For the problem of Xen original memory migration algorithm making migration low efficient, when the virtual machine dirty page rate is high, we propose a improved pre-copy algorithm of weighted statistic, which adds up weighted dirty pages information collected, selects and transfers pages that are modified in low frequency and are not being used in a period of time preferentially. The pages that are modified frequently are transferred after virtual machine is suspended. This algorithm can effectively reduce the amount of memory pages transferred and the total migration time.Secondly, we design and implement a migration management framework adapting to cloud computing, which can move virtual machine by manual control and automatic control. In automatic migration model, the migration management framework can detect the load state of physical machine through load information of monitoring system. When a physical machine’s resource is used overload and become a hotspot, the migration management framework will automatically select part of virtual machines on that physical machine and migrate to another, to reduce resource utilization and eliminate hotspots.Finally, through a large number of experiments, we analyze and verify the performance of pre-copy algorithm of weighted statistic proposed and the function of migration management framework. Experiment results show that when virtual machine under heavy load, using the algorithm proposed, total migration time can be shortened by15%, and the total pages transferred can be reduced by20%, with downtime not increasing obviously. Form the view of memory usage of physical machine we verify the function of migration management framework. Results show that the proposed migration framework can detect memory usage hotspot timely, and eliminate hotspot to achieve load balance automatically through migrating virtual machines dynamically.