SR-PSO: server residual efficiency-aware particle swarm optimization for dynamic virtual machine scheduling

A dynamic virtual machine scheduling is the discrete optimization problem that schedules virtual machines over the set of physical servers at each discrete scheduling interval. As this problem is NP-complete, heuristic and greedy approaches may get stuck at the local minima and produce the suboptimal solution. Therefore, we proposed server residual efficiency-aware particle swarm optimization (SR-PSO) algorithm for dynamic virtual machine scheduling in this work. The classical PSO operators are tuned to suit dynamic virtual machine scheduling. The proposed bi-objective fitness function guides the proposed algorithm during the exploration of global solution space and schedules virtual machines over the physical servers operating at optimum energy efficiency or near it with minimum virtual machine migrations. A virtual machine selection algorithm is proposed that selects the virtual machines whose migration results in servers’ optimum energy efficiency. The server underload detection algorithm is proposed that categorizes servers as underloaded if they operate with energy inefficiency. The SR-PSO algorithm is aware of discrete scheduling intervals, and at each scheduling interval, only those VMs are rescheduled that are prone to service level agreement SLA violation or lower server utilization. We have used a cloudsim simulator to simulate our proposed work, and the results show significant improvement in energy consumption for the dynamic VM scheduling. More specifically, our proposed approach is 45.4% and 50% more energy efficient than the previous dynamic virtual machine scheduling approaches.