Energy-Efficient VM Scheduling in the Cloud Environment using Reinforcement Learning

Abstract

Cloud data centers consume a huge amount of energy in the form of electrical energy for their operation. They also emit carbon dioxide and impact the balance of nature. This management of exponentially increasing load and the minimization of energy use along with the impact on the environment is the biggest challenge a cloud service provider (CSP) faces. CSPs establish and maintain data center farms, which enable the delivery of cloud services to millions of clients. The reduction in energy usage by data centers while also minimizing the number of service level agreement (SLA) violations is a major challenge. In this work, we have proposed a reinforcement learning (RL)-based dynamic virtual machine (VM) consolidation mechanism wherein the host load is predicted by considering previous and current host utilization. The learning agent chooses a suitable-power mode for the hosts. Load balancing is done for the over-utilized hosts and dynamic VM consolidation is performed for the under-utilized hosts. The VM scheduling is performed using an energy-aware best fit method. Ourproposed model shows a significant drop in the number of SLA violations and energy consumption when compared to the ARIMA model.