Switch migration frequency for load balancing in fog computing using machine learning algorithm

Abstract

With the daily rise in Internet usage due to affordable smart devices and easy access to advanced technologies, there is an increased strain on network cores, particularly from the Internet of Things (IoT) traffic. Although this advancement aids in simplifying daily tasks, it leads to network congestion, demanding more robust network infrastructures. A promising solution lies in fog computing, a decentralised approach that brings cloud services closer to the network edge, aiming to enhance efficiency by reducing data transferred to the cloud. This solution boosts efficiency and addresses security and compliance concerns, especially with the exponential data challenges associated with IoT. Despite the benefits, an increase in connected devices results in a surge in network packets, potentially overburdening the fog environment. A viable response to this challenge is load balancing among fog nodes, mainly using switch migration, redirecting traffic from overloaded to underloaded controllers. However, frequent migrations may affect network latency and service quality. Implementing machine learning can potentially reduce migration frequency by forecasting traffic patterns. The experimental results suggest that the specific traffic characteristics within the network determine the need to minimise migration. In constantly high-traffic situations, a simple threshold approach is practical. At the same time, machine learning can optimise networks with variable traffic, with an 87%-100% reduction in migration frequency, avoiding unnecessary costs that impede performance.