Saving time and cost on the scheduling of fog-based IoT applications using deep reinforcement learning approach

Due to the rapid growth of intelligent devices and the Internet of Things (IoT) applications in recent years, the volume of data that is generated by these devices is increasing ceaselessly. Hence, moving all of these data to cloud datacenters would be impossible and would lead to more bandwidth usage, latency, cost, and energy consumption. In such cases, the fog layer would be the best place for data processing. In the fog layer, the computing equipment dedicates parts of its limited resources to process the IoT application tasks. Therefore, efficient utilization of computing resources is of great importance and requires an optimal and intelligent strategy for task scheduling. In this paper, we have focused on the task scheduling of fog-based IoT applications with the aim of minimizing long-term service delay and computation cost under the resource and deadline constraints. To address this problem, we have used the reinforcement learning approach and have proposed a Double Deep Q-Learning (DDQL)-based scheduling algorithm using the target network and experience replay techniques. The evaluation results reveal that our proposed algorithm outperforms some baseline algorithms in terms of service delay, computation cost, energy consumption and task accomplishment and also handles the Single Point of Failure (SPoF) and load balancing challenges. (C) 2019 Elsevier B.V. All rights reserved.