A Load Balancing and Routing Strategy in Fog Computing using Deep Reinforcement Learning

Fog computing is proposed to overcome cloud computing limitations by extending its services close to the network edge. However, fog systems are still facing many challenges due to their large-scale distributed architecture and small resource power. We investigate in this paper resource management in fog computing and propose a load balancing strategy using Deep Reinforcement Learning (DRL) combined with a link-state routing protocol based on the Dijkstra algorithm. The objective of the strategy is to simultaneously minimize the tasks’ processing and communication delay. The proposed solution involves designing a Load Balancer Smart Controller (LBSC) which engages a smart DRL agent in a fog environment. The LBSC examines the nodes and links’ states in order to make the optimal decision by selecting the most suitable node and path to process each task. The performance of the proposed approach is tested in a dynamic IoT environment with a high-rate workload generation scenario. Simulation results show that the average total latency including processing and communication delays of the proposed method is reduced by around 50% in comparison to the classic load balancing algorithms.