Research On Workload-aware Virtual Machines Initial Placement And Migration Decision

Cloud Data Center (DC) is the center of data operation, exchange and storage. Based on virtualization technology, virtual machine(VM) placement has become an important technology for power management and elastic resource provision. In the stage of initial VMs placement, depending on the size of workload in DC, VMs placement mainly adopt greedy or heuristic initialization strategy.In the stage of dynamic VMs management, with the changes of resources utilization, migration trigger mechanism will determine when to migrate the VMs from one host to the other. The accurate judgment of trigger time can balance the hot spots effectively and turn off cold spots in DC. However, current migration trigger mechanism lacks the response to the changes of DC workloads, and static threshold configurations are easy to cause frequent migration with unnecessary migration and transmission cost.In order to solve the above problems, the main contributions of this dissertaton are as follows:(1) This dissertation proposed workload-aware merge placement algorithm(WAM) for different types of workload in initial VM placement, which is different from existing algorithms. Compared with different initial placement methods(WAM, BF, RR), the experiment results show that WAM can effectively save power consumption, balance workload, and reduce the number of used physical machines.(2) This dissertation proposed a dynamic threshold setting algorithm iWnd to decide when to migrate VMs. Different from existing algorithms, iWnd can adjust the size of threshold window according to the whole workloads in DC. In this way, iWnd can reduce the number of VMs which need to be migrated, avoiding unnecessary migration and transmission cost and saving power. Experiments show that iWnd can effectively reduce the number of VMs migration and migration failure rate without producing additional power consumption.(3) In order to find the effective association between initialization placement and dynamic management, we combine WAM and iWnd algorithms, this dissertation proposed workload-aware VMs migration trigger decision. In this mechanism, we analyse impact of migration timing when considering different types of VMs. The experimental results show that WAM_iWnd make load balance in DC and reduce power consumption effectively, and does not result in a lot of extra migration cost.