RLPRAF: Reinforcement Learning-Based Proactive Resource Allocation Framework for Resource Provisioning in Cloud Environment

Abstract

Recent developments in cloud technology enable one to dynamically deploy heterogeneous resources as and when needed. This dynamic nature of the incoming workload causes fluctuations in the cloud environment, which is currently addressed using traditional reactive scaling techniques. Simple reactive approaches affect elastic system performance either by over-provisioning resources which significantly increases the cost, or by under-provisioning, which leads to starvation. Hence automated resource provisioning becomes an effective method to deal with such workload fluctuations. The aforementioned problems can also be resolved by using intelligent resource provisioning techniques by dynamically assigning required resources while adapting to the environment. In this paper, a reinforcement learning-based proactive resource allocation framework (RLPRAF) is proposed. This framework simultaneously learns the environment and distributes the resources. The proposed work presents a paradigm for the optimal allocation of resources by merging the notions of automatic computation, linear regression, and reinforcement learning. When tested with real-time workloads, the proposed RLPRAF method surpasses previous auto-scaling algorithms considering CPU usage, response time, and throughput. Finally, a set of tests demonstrate that the suggested strategy lowers overall expense by 30% and SLA violation by 77.7%. Furthermore, it converges at an optimum timing and demonstrates that it is feasible for a wide range of real-world service-based cloud applications.