ESPS: Energy Saving Power Spectrum-Aware Scheduling to Leverage Differences in Power Ratings of Physical Hosts in Datacenters

摘要

Cloud Computing has seen massive growth over the past couple of decades, leading to exponential growth in energy consumption at data centres. Data centres consuming high amounts of energy leave a carbon footprint of the same scale, hence Cloud Service Providers (CSPs) have been looking for energy-efficient solutions to task scheduling in cloud to reduce the amount of carbon dioxide emission. Saving energy not only helps reduce the carbon footprint datacentres have on the environment, but also helps cover the costs of running multiple datacentres on the CSP’s end. In this paper, we propose an energy saving task scheduling heuristic for heterogeneous cloud systems which selects the optimal physical host containing virtual machines with the additional consideration of the utilization of any incoming task on that particular virtual machine. We compare the energy efficiency of our proposed heuristic with recent algorithms including ECTC, MaxUtil, Random, and FCFS on several benchmark and synthetic datasets to display its superiority in energy-efficient task scheduling in heterogeneous cloud environments. FCFS, MaxUtil, Random, and ECTC respectively consume approximately 38.65%, 33.59%, 53.02%. and 46.96% more energy in a heterogeneous cloud environment as compared to our proposed heuristic namely Energy Saving Power Spectrum-Aware Scheduling (ESPS).