Research On Live Migration Of Virtual Machines And Load Balance In Cloud Computing

Cloud computing allows hosting multiple services on the shared resources pool,and allocate them on demand. Virtualization is the important way to improve theflexibility of provisioning the resources. This is done by dynamically changing theallocation of virtual machine resources and migrating virtual machines across thephysical servers. Live migration of virtual machines is an important tool for themanagement of cloud computing. For example, with the live migration, the load onthe congested host will be relieved or by moving a virtual machine to an alternativeone for the maintenance. While live migration of virtual machines brings versatility,but the performance will go down during the migration, so shorter downtime and totalmigration time is highly desirable.Combined with Xen virtual machine which is widely used now and pre-copymigration mechanism, proposes the live migration strategy of Xen virtual machine,specific works are as follows:First, before the memory migration of the virtual machine, the virtual machinemonitor selects the destination host and establishes connections. Traditional loadbalance algorithms consider the single load balance evaluation, such as theleast-connection scheduling algorithm only records the request connections. However,it is able to represent the load of every node, but computing resources, such as CPU,occupied by each connection may be different. For its own characteristics of cloudcomputing, proposes destination node locating strategy based on live migration ofvirtual machines, periodically collecting the usage of CPU, memory, bandwidth andthe number of request connections, then using the same unified standard to quantifythem, considering the migration distance between the source host where virtualmachine located and the destination host, computing the integrated value and settingthe value as the selecting criteria. This method measures the utilization of computingresources and migration costs, and it can be reasonable and efficient to locate thedestination node.Second, Xen uses pre-copy to migrate memories, i.e. the pages to be transferredduring round n are those that are modified during round n-1, this step performs initerative, until dirty pages is less than the threshold, then is the stage of stop-and-copy. For multiple iterative copies of memory pages, proposes optimized memory migratingalgorithm based on probability prediction, depending historical records, calculatingthe probability of each memory page is modified before the next round and then basedon the probability value, transferring memory pages in order. The optimizedalgorithm reduces the amount of memory pages each round transferred, therebyshortening the total migration time, to some extent, improving the performance ofmigration.Finally, for destination node locating strategy based on live migration of virtualmachines and optimized memory migrating algorithm based on probabilisticprediction, we have experiments under different application scenarios, recordingexperimental data, analyzing the utilization of CPU, memory, bandwidth anddowntime, total migration time, the amount of transferred memory pages. The resultsshow that our algorithm can ensure the system load balance while reducing totalmigration time, thereby increasing the efficiency of the migration.