Computer Science · Computer Networks and Communications
How Can the Integrated Routing and Scheduling Enhance Optimality Bounds of Time-Sensitive Transmission
Zhenrui Cao, Tie Qiu, Qingyong Deng, Haolin Liu, Xiaobo Zhou
See on the mapAbstract
Time-sensitive networking (TSN) ensures reliable real-time transmission services for diverse applications in the Industrial Internet of Things through strictly deterministic time-triggered (TT) flows. The end-to-end transmission performance of TT flows is jointly influenced by routing choices and bandwidth scheduling, making their optimization essential for improving the overall service quality of TSN networks. Existing methods address bandwidth scheduling only after determining the route, leading to a disjointed approach to routing and scheduling decisions. This separation may introduce potential transmission conflicts, such as routes with insufficient available bandwidth being selected prematurely, resulting in the failure to allocate the required bandwidth resources for deterministic flow transmission. To maximize the transmissibility of TT flows, we propose MAIRS, a multi-agent deep reinforcement learning (MADRL)-based Integrated Routing and Scheduling method. First, leveraging network softwarization, we design a synchronized decision-making framework for integrated routing and scheduling to enhance the optimality bounds of TT flow transmission. Next, we introduce collaborative MADRL to develop comprehensive end-to-end transmission plans for TT flows, effectively addressing the intricate integrated routing and scheduling problem characterized by a vast solution space. Experiments demonstrate that MAIRS significantly outperforms existing algorithms in improving the transmission performance of TT flows.