Reinforcement-Learning-Based Software-Defined Edge Task Allocation Algorithm

Abstract

With the rapid growth in the number of IoT devices at the edge of the network, fast, flexible and secure edge computing has emerged, but the disadvantage of the insufficient computing power of edge servers is evident when dealing with massive computing tasks. To address this situation, firstly, a software-defined edge-computing architecture (SDEC) is proposed, merging the control layer of the software-defined architecture with the edge layer of edge computing, where multiple controllers share global information about the network state through an east–west message exchange, providing global state for the collaboration of edge servers. Secondly, a reinforcement-learning-based software-defined edge task allocation algorithm (RL-SDETA) is proposed in the software-defined IoT, which enables controllers to allocate computational tasks to the most appropriate edge servers for execution based on the global network information they have obtained. Simulation results show that the RL-SDETA algorithm can effectively reduce the finding cost of the optimal edge server and reduce the task completion time and its energy consumption compared to various task allocation methods such as random and uniform.