DQRL: A Directed Acyclic Graph and RL-Based Framework for QoS-Centric Routing in Multi SDNs

Abstract

The growing complexity of modern networks demands adaptive and scalable traffic management to satisfy diverse Quality of Service (QoS) requirements. While Software-Defined Networks (SDNs) provide programmability and flexibility, traditional routing algorithms such as OSPF and Dijkstra struggle to respond to dynamic conditions. This study proposes a novel QoS-driven routing framework, DQRL, integrating Directed Acyclic Graph (DAG)-based Distributed Ledger Technology (DLT) and Reinforcement Learning (RL) to enable decentralized and adaptive routing. DQRL employs DAG-based DLT as a decentralized ledger to maintain distributed and up-to-date routing information, eliminating centralized dependencies. Unlike blockchain, DAG supports parallel transaction validation, reducing latency and enhancing scalability for real-time networks. RL, particularly Q-learning, dynamically selects optimal paths using QoS metrics like bandwidth, delay, and packet loss, ensuring resilient inter-AS routing. Experimental results demonstrate that DQRL improves QoS-aware routing performance while reducing control overhead. The findings highlight the potential of combining DAG-based DLT with RL to meet the challenges of inter-domain SDN routing. © 2025 IEEE.