Distributed SDN control for IoT networks: A federated meta reinforcement learning solution for load balancing

Abstract

The growth of Internet of Things devices and their uses have introduced ample challenges in handling dynamic and heterogeneous traffic patterns. This also has affected the area of Software Defined Networking (SDN). The key parameters like scalability, latency and resilience are the concerns in centralized SDN approach, especially in the case of large-scale IoT deployments. This research introduces a new method, Distributed SDN Control for IoT networks: A Federated Meta Reinforcement Learning Solution for Load Balancing. This method combines Federated Learning (FL) with the key features of Meta Reinforcement Learning (Meta-RL) to enable intelligent and privacy preserving load balancing across distributed SDN controllers. The system functions in two phases. In the first phase, traffic distribution models across are trained with FL without sharing raw data. Security is added to this by differential privacy and Byzantineresilient aggregation. In the second phase, fast adaptation to non-stationary traffic patterns is achieved using Meta-Learning and Proximal Policy Optimization (PPO). The performance evaluations show that the