DDSPQ: Strict-Priority Queue Schedule for Multi-Tiered Service Accommodation in SD-WAN

Software-defined wide area networks (SD-WAN) decouple network control from hardware, enabling flexible and application-aware traffic management. However, SD-WANs are designed with a distributed control plane to ensure scalability, while traditional queue scheduling mechanisms exhibit limitations in effective collaboration between controller. To address these challenges, we propose a DGN-based dynamic strict-priority queue (DDSPQ) schedule framework that ensures compliance with maximum average latency and minimum throughput requirements for multi-tiered services. Our solution integrates a graph convolutional reinforcement learning (namely DGN) based algorithm to assign priorities for flows through localized agent interactions, eliminating global synchronization overhead. Additionally, we introduce a priority-queue mapping algorithm to extend the number of priorities under insufficient hardware while prioritizing mission-critical flows. The proposed framework is implemented with P4-programmable data planes and evaluated on an SD-WAN topology. Experimental results indicate that our solution can effectively satisfy the differentiated multi-tiered quality of service requirements with low communication overhead.