Meta-Reinforcement Learning Based Resource Management in Software Defined Networks Using Bayesian Network

The capacity of Software Defined network (SDN) to efficiently manage resources depends heavily on the scalability of both the scale of the network and the services it supports. An effective resource allocation (RA) method is necessary for managing dynamic network traffic and load distribution across several controllers. The uncertain and dynamic relationship between resources has prevented reinforcement learning (RL) from performing effectively for real-time load in SDN, despite its use for load balancing. This is because reinforcement learning models resource management as a linear optimization issue. This paper presents a Bayesian framework to build an intelligent optimization framework for SDN resource management by employing deep meta-RL. It is a variant of conventional RL that requires less data for training but still helps the agent grasp the underlying policies. To accomplish load balance, the Bayesian network is trained with RL to make the most optimal decisions based on the predicted amount of congestion. Adjusting the controller’s parameter weights automatically helps it cope with congestion caused by heavy loads. This algorithm decides on the best action based on the prediction made by reinforcement learning. In this case, the resource management strategy proposed by SDN has been experimentally validated to confirm the theoretical understanding. This work used a real-time database from the electricity distribution center in Indore, India data center to assess the proposed work’s efficiency and efficacy.