SDN-Enabled Adaptive and Reliable Communication in IoT-Fog Environment Using Machine Learning and Multiobjective Optimization

The Internet-of-Things (IoT) devices, backed by resourceful fog computing, are capable of meeting the requirements of computationally-intensive tasks. However, many existing IoT applications are unable to perform well, due to different Quality-of-Service (QoS) requirements, while communicating with the fog server. Besides, constantly changing traffic demands of applications is another challenge. For example, the demand for real-time applications includes communicating over a path that is less prone to delay, and applications that offload computationally intensive tasks to the fog server need a reliable path that has a lower probability of link failure. This results in a tradeoff between conflicting objectives that are constantly evolving, i.e., minimizing end-to-end delay and maximizing the reliability of paths between IoT devices and the fog server. We propose a novel approach that takes advantage of machine learning (ML) and multiobjective optimization (MOO)-based techniques. The reliability of links is evaluated using an ML-based algorithm in an software-defined network (SDN)-enabled multihop scenario for the IoT-fog environment. By considering the two conflicting objectives, the MOO algorithm is used to find the Pareto-optimal paths. Our experimental evaluation considers two applications with different QoS requirements-a real-time application (App-1) using UDP sockets and a task offloading application (App-2) using TCP sockets. Our results show that: 1) the tradeoff between the two objectives can be optimized and 2) the SDN controller was able to make adaptive decision on-the-fly to choose the best path from the Pareto-optimal set. The App-1 communicating over the selected path finished its execution in 13% less time than communicating over the shortest path. The App-2 had 41% less packet loss using the selected path compared to using the shortest path.

Automated summary

IoT devices with fog computing can handle computationally-intensive tasks, but face challenges in meeting different QoS requirements and changing traffic demands when communicating with fog servers. A proposed approach utilizes machine learning and multiobjective optimization to evaluate link reliability and find Pareto-optimal paths.