Research On The Migration Of Virtual Machines In Cloud Computing

Virtual machine technology can virtualize hardware resources, which can let many different virtual machines running on a single physical host. These virtual machines run on the isolated virtual hardware environment. Although they may have different operating systems, they can provide same services as real physical hosts. Virtual machine technology improves the utilization of the host hardware, fulfills the load balancing and simplifies the management work.Live migration of virtual machines is an important feature of the virtual machine technology. Live migration can migrate a virtual machine from one host to another host completely and uninterruptedly while the machine is still running.Two important indicators for the evaluation of live migration are the total migration time and the downtime. Currently the dynamic bandwidth migration algorithm proposed by Clark is the most commonly used algorithm in live migration. The algorithm is used by the open-source virtual machine monitor Xen, which can reduce the downtime to the millisecond. However, the clark algorithm cannot avoid the transmission of higher frequency change memories during the migration process, and also cannot reduce the speed of memories become dirty. This will reduce the overall efficiency of virtual machine migration.This article describes the background and development of the technology of the virtual machine, briefly introduce the Xen virtual machine as well as its implementation and introduce the principle and achieve of dynamic migration of virtual machines. Then analyze the source code of the Xen virtual machine migration part, pointed out the lack of Writable Working Set (WWS) judgments algorithm, and propose an optimized solutions to avoid the inadequacies. The article also analyzes the CPU scheduling algorithm of Xen, which is Credit algorithm. And propose an improved deceleration scheduling algorithm from the aspect to reduce dirty page generation. At last, we compare the two improvement algorithms and the original proposal through simulation and experiment. As can be seen in the result, the two improvement algorithms further shortens the total migration time and downtime than the original one.