Research On Dynamic Resource Configuration Based On Virtual Machine Migration

System virtualization effectively tackles the important questions such as severe resource waste, enormous management costs in traditional cluters or data centers. In virtualized environments, the basic problem is dynamic resource reconfiguration due to time-varying workloads. Virtual machine migration is used to remapping resources to applications. This dissertation performs a detailed study on the dynamic remapping of VMs to physical nodes in terms of the optimization of VM migration, server consolidation, removing hot computing nodes.First of all, virtual machine migration is one of the most important approaches of dynamic resource scheduling in virtualized environments. The existing methods of VM migration have enormous transferred data and consume long time. Using page compression, a new VM migration approach in an adaption mode is presented to automatically adapt to different network environment to keep the improved performance of live VM migration. Operations save and restore is also a way to migrate VMs. A fast saving/restoring VM method is proposed to improve the performance of network file system using page compression. Different compression methods are used to compress pages with different characteristics to better balance the relationship of compression ratio and compression rate. Experimental results show that, operations save and restore using page compression reduce time respectively by69.4%and62.0%on average. The size of memory image file is shrinked by83.1%at the most.Server consolidation is the main approach for green computing in data centers. Most consolidation methods use heuristics to find a locally-optimal solution. A linear-programming-based algorithm is presented to find an optimal solution using constrains about the relationship between resource demands and supplies. The creation of VM migration order first identifies the time dependency relationship between some VMs. Then, it uses parallel migrating mode and VM migration based on page compression to decrease the costs of resource reconfiguration. The results of simulation is that, the given consolidation algorithm can effectively reduce the number of active physical nodes by26.7%on average.Removing hot nodes is one basic problem of dynamic resource configuration in virtualized environments. When the workloads increase, resource contention is aggravated. The remapping of VMs to nodes is done to remove hot nodes. Removing hot nodes includes two sub-problems:one is to select a sub-set from the VM set on a hot node, namely VM selection problem; the other is to confirm target nodes for migrated VMs, namelyVM placement problem. For VM selection problem, the existing methods give a simple cost estimation based on the size of memory. A real cost prediction approach based on pre-copy is proposed. It first analyzes the process of pre-copy algorithm and finds the factors affecting the migration cost. Then, it creates the relationship of the application characteristics and costs and forms the migration cost estimation model.For VM placement problem, a workload-aware resource reconfiguration is presented based on history data. When redistributing virtual machines among physical nodes, the new remapping not only is based on current the resource demands, but also takes the variation trend of the workloads in the near future into accounts. According to the prediction of the workloads, the new mapping should both meet the resource demands of virtual machines in the present and in the near future, which finally avoids the frequent migration of virtual machines. The results of simulation experiments show that, comprared to FFD (First-Fit Decrease), the given approach reduce the total migration time by75%.