Balancing Resource Utilization for Continuous Virtual Machine Requests in Clouds

The placement of virtual machines (VMs) on a cluster of physical machines (PMs) is a primary task in clouds. We can benefit a lot from appropriate placement policy, e.g. cost saving. In this paper, we raise a continuous virtual machine placement problem and propose an on-line algorithm to reduce the power consumption of clouds. Generic methods which deal with this problem with greedy manner will result in resource leak, a phenomenon of resource wasting. The direct consequence of this phenomenon is larger number of running PMs, which will cause higher power consumption. We prove it is an NP-complete problem to minimize the power consumption during the placement of VMs while the VM requests reach the clouds continuously. We propose a balanced algorithm with the principle of avoiding resource leak. The balanced algorithm handles the VM requests in real time and guarantees high resource utilization. The number of PMs and power consumption can be cut down as a result of high resource utilization. We evaluate the balanced algorithm via experiments with various simulations. Experimental results show that the balanced algorithm can reduce the number of running PMs effectively, so that, it has a significant efficiency on power saving.