Efficient Adaptive Resource Provisioning for Cloud Applications using Reinforcement Learning

Abstract

An appealing feature of cloud computing is elasticity, that allows shrinking or expanding the resources allocated to an application in order to adjust to workload variations. The resource provisioning algorithm must also adhere to the performance requirements specified in the Service Level Agreement between the cloud provider and the client who runs the application. While the use of Reinforcement learning algorithms such as Q-learning has been proposed already to address this problem, those suffer from slow convergence and scalability issues. In this paper, we explore methods for overcoming such challenges and ensuring effective resource utilization. Preliminary experiments on CloudSim platform demonstrate the superiority of some of these methods over static, threshold-based and other reinforcement learning based allocation schemes.