Exploring Teacher-Student Learning with Multi-Agent DRL for QoS Routing in SDN

The emergence of Software-Defined Networking (SDN) has marked a profound shift in the landscape of network management, revolutionizing how networks are designed, operated, and controlled. However, the task of attaining optimal routing in SDN while enhancing efficient network performance and ensuring the highest Quality of Service (QoS) still remains a challenging problem. This paper introduces a novel Distributed Teacher-Student (DTS) method that jointly harnesses the power of Multi-Agent Deep Reinforcement Learning (MADRL) with Knowledge distillation for effective QoS routing. The core idea is to transfer knowledge from a domain expert (the teacher) to an intelligent agent (the student) through a distributed learning process using DRL. This collaboration involves multiple teacher and student entities working together to decide and create optimal routes while considering the specific QoS requirements for each type of data flow. The experimental findings vividly demonstrate the effectiveness of the proposed DTS framework, revealing significant enhancements in network performance, including delay and throughput ratio. Compared to the conventional MADRL approach without the Teacher-Student framework, our DTS solution showcases a remarkable improvement of 30% in normal traffic scenarios and over 52% in situations with link failures.