Reinforcement Learning Techniques for Optimizing System Configuration on the Cloud: A Taxonomy and Open Problems

Efficient resource management (RM) is paramount for achieving high performance and utilization of computing resources in cloud computing environments. Conventional approaches, such as rule-based heuristics and optimization algorithms, face challenges in adapting to the dynamic and intricate nature of these environments. In this work, we investigate the utilization of reinforcement learning (RL) techniques for RM on the cloud. We provide a comprehensive taxonomy that categorizes RL-based approaches according to various facets of RM, encompassing resource allocation, auto-scaling, load balancing, and energy efficiency. By conducting an extensive literature review, we analyze and compare diverse RL algorithms employed in RM, highlighting the strengths and limitations of each approach. Last, we identify potential research directions in the context of RL-based resource management methods on the cloud.