Design and implementation of an efficient load-balancing method for virtual machine cluster based on cloud service

With the expansion of cloud-based services, the question as to how to control the workload among a large virtual machine (VM) cluster has become increasingly important. One of the most popular way to achieve load- balancing is migrating VMs from high-load machine to light- load ones. But the migration of VMs has a large granularity which must be considered. In this paper we addresses the problem with a policy for creating a load-balancing method in both PMs layers and VMs layers, moreover we also introduced prediction method to ensure the transient spike does not trigger needless VMs migration, and in VMs layers we also purpose a benefit estimates model in order to decide whether the migration of jobs in VMs under the same PM is benefit for the whole system. The evaluation results we provided showing that substantially the dynamic reallocation of VMs brings load-balancing.