Mvmotion: A metadata based virtual machine migration in cloud

As one of the most important features of virtualization, virtual machine (VM) migration provides great benefits for load balancing, resources-saving, fault tolerance in modern cloud data centers. Considering the network traffic caused by transferring data during VM migration imposes a huge pressure on network bandwidth of cloud data centers, and by analyzing the characteristic of the transferred data, we found that the redundant data, which is produced between two physical hosts by hosting virtual machines cloned from same VM template, can be reduced to relieve the network traffic pressure. This paper presents a Metadata based VM migration approach (Mvmotion) to reduce the amount of transferred data during migration by utilizing memory de-redundant technique between two physical hosts. Mvmotion utilizes the hash based fingerprints to generate Metadata of memory, which is used to identify redundant memory of VMs between two hosts. Based on the Metadata, the transfer of redundant memory data during migration can be eliminated. Experiment demonstrates that, compare to Xen's default migration approach, Mvmotion can reduce the total transferred data by 29-97 %, and decreases the migration time by 16-53 %.