Multi Agent Energy Efficient Model for Cloud Resource Management and Consolidation Using Reinforcement Learning

Abstract

Cloud Computing provides services to the customer by using various resources which consumes energy. Energy plays a vital role in the cloud because it involves cost and service quality. Various models are implemented for reducing the consumption of energy. Existing models analyzed only a specific set of resources that are not efficient because of the large number of resources involved in the data center. Reinforcement Learning (RL) model gives support to minimizing the energy through learning and reward. Single RL models also analyze the energy in resource level. The Data Center (DC) in the cloud consists of a host with a number of Virtual Machines (VMs), so the energy consolidation is needed for minimizing energy. Multi agent RL model is proposed for minimizing the energy in VM level to DC level by implementing three agents namely VM agent, Host Agent and DC agent. The reward value is calculated based on these agents with corresponding resource level. It performs the action related to policy for finding energy efficient resources in order to execute the workload. Idle VM on the host are identified and terminate it because it consumes a substantial amount of energy. The proposed RL model consolidates the VMs of the various host and shutdown idle hosts and achieves minimum energy.