Abstract
The existence of distributed denial of service (DDoS) attacks can lead to link failures, node congestion,and routing table changes, causing dynamic changes in network topology, which requires routing design to be dynamically adaptable and fault-tolerant. To this end, we propose a Graph neural network (GNN)-enabled multi-agent Deep reinforcement learning (DRL) Adaptive Path Selection scheme, termed GDAPS. Specifically, we construct a hierarchical Software-defined networks (SDN) paradigm to manage multi-network domains and support parallel routing path calculation. The proposed GNN algorithm is used to model the topological relationship between nodes and links, and capture the dynamic changes of the network topology to update the network state in real time. Then, we propose a hierarchical collaborative multi-agent DRL algorithm to improve the stability and efficiency of network services in dynamic multi-network domain. Especially, we use GNNs as an effective enabler for DRL agents to understand and adapt to network topology and dynamic changes, ensuring that DRL agents can dynamically adjust path strategies in real-time according to changes in network status. Finally, compared with other baseline methods, the experimental results show that our proposed GDAPS scheme has optimal performance in terms of delay, packet loss rate, and throughput when the network state dynamic changes.