MARS: Measurement-based Allocation of VM Resources for Cloud Data Centers

High performance data centers use virtualization technique which enables each physical server machine to host multiple virtual machines (VMs) to achieve highly efficient resource utilization. In this paper, we propose a measurement-based approach for efficient allocation of virtualized resources in hyper-convergence environments where virtualized computing, networking, and storage resources are unified and converged. Using real-time measurements of service performance metrics, our proposed approach identifies the VM with the worst performance resulting from over-utilized resource, and gradually adjusts the amount of resources allocated to it in order to improve its performance. The results of empirical evaluations conducted indicate that our proposed approach can realize efficient resource allocation among VMs with varying resource demands.