MFF: Performance Interference-Aware VM Placement Algorithm for Reducing Energy Consumption in Data Centers

Virtualization is the main technology that powers cloud computing and has enabled the execution of multiple applications in same physical hardware using virtual machines (VM) for efficient utilization of resources and energy savings. Although virtualization successfully isolates co-resident VMs from a security perspective, it does not offer a guarantee from a performance interference perspective. This means that sharing of resources results in competition, which is the cause of performance interference. Performance interference is more pronounced in homogenous workloads, where applications workloads contend to the same shared resource. In this case, application workloads run for longer times due to reduced performance and thus consume more energy. To address this problem, a VM allocation policy should ensure that VM running homogeneous workloads is not co-located. In this paper, we propose a VM allocation algorithm called Minimum Interference First Fit (MFF), which co-locates dissimilar workloads. The algorithm clusters VMs using K-means based on resources usage. Before a VM is placed into a physical machine (PM), similarity index (SI) of all the active PMs is computed, the VM is then placed in a PM with least SI. MFF has been evaluated on a simulated data center using CloudSim Plus cloud simulator on application workloads logs obtained from a production data center. Results show that MFF outperforms well-known VM allocations algorithms such as first fit (FF), worst fit (WF) and best fit (BF) from an energy consumption perspective.