Joint Optimization of Energy Consumption and Latency Based on DRL: An Edge Server Activation and Task Scheduling Scheme in IIoT

Edge computing has been proposed as a promising solution to alleviate the computation intensive requirement of Industrial Internet of Things (IIoT) scenarios. In edge computing based network, task latency and energy consumption are two key metrics, while the tradeoff of them is of great importance on impacting the overall performance of the system. In this paper, we formulate a joint optimization problem to minimize the weighted summation of latency and energy consumption in the network where the task scheduling and server dormant mode are both taken into account. To solve this problem, we designed a Deep Reinforcement Learning (DRL) based algorithm considering both the number of active edge servers and the task scheduling scheme per time slot. Simulation results show that our algorithm has advantages compared with other algorithms and reduces the overall cost of the system.