QoS-Driven Distributed Cooperative Data Offloading and Heterogeneous Resource Scheduling for IIoT

Edge computing has become a powerful paradigm to fulfill the diversified quality of service (QoS) demands of the Industrial Internet of Things (IIoT) applications. This study examines the cooperative data offloading (DO) and heterogeneous resource scheduling (RS) problem for maximizing long-term system utility. Owing to the dynamics, high connectivity density, and diverse QoS demands of IIoT, a QoS-driven distributed decision-making (QDDM) framework is proposed to address this problem. Specifically, this framework decomposes the primal problem into two subproblems: industrial terminal device (ITD)-side DO and edge server (EDS)-side RS. Then, a modified soft actor-critic (SAC)-based multi-agent deep reinforcement learning (MSMD) algorithm is proposed to address the ITD-side DO subproblem, which can achieve more accurate estimation of the Q-values and solve both the centralized-decentralized mismatch and the multi-agent credit assignment issues. Based on the DO decisions of each ITD, a linear approximation method is proposed to transform the EDS-side RS subproblem into an easily-solved linear programming subproblem. Finally, a real-world IIoT experiment platform is built to evaluate the performance of the QDDM framework. The evaluation results demonstrate that the QDDM framework effectively increases the long-term system utility.