Intelligent dynamic bandwidth allocation for real-time IoT in fog-based optical networks

The rapid proliferation of Internet of Things (IoT) applications has led to an exponential increase in data traffic, which is imposing a significant burden on existing network infrastructures. Ethernet Passive Optical Networks (EPONs) within fog computing environments are critical for supporting real-time IoT applications because of their high bandwidth and low-latency capabilities. However, traditional static bandwidth allocation methods designed for EPONs do not accommodate the dynamic and heterogeneous nature of real-time IoT traffic, leading to inefficient resource utilization and service degradation. This paper proposes a reinforcement learning (RL)-driven fog-based dynamic bandwidth allocation (DBA) framework to optimize resource management in EPONs with multi-optical edge devices (OEDs) within cloud-fog computing environments. The framework improves bandwidth distribution for OEDs connected to IoT access points through a fog server, ensuring adaptive real-time dynamic bandwidth allocation. Specifically, this paper formulates the bandwidth allocation problem as a constrained profit-maximization discounted return problem, which cannot be directly solved using traditional optimization methods due to the dynamic and uncertain nature of the IoT traffic. To effectively solve this problem, this paper models the bandwidth allocation problem as a Markov Decision Process (MDP), which can be solved using RL without requiring prior traffic knowledge. Through a continuous real-time learning process, the proposed algorithm learns the IoT traffic dynamics and then adaptively optimizes bandwidth allocation at the fog layer, resulting in improved system efficiency and adaptability. Simulation results demonstrate that, compared to conventional bandwidth allocation algorithms designed for fog-based EPONs (i.e., equal distribution and moving average algorithms), the proposed RL-based framework significantly improves bandwidth utilization and reduces blocking probability related to IoT networks by dynamically adjusting bandwidth allocation based on the learned real-time traffic.