TFIR: GCN Powered and Reinforcement Learning-based Topo-Free Intelligent Routing

The development of the Internet has given rise to a variety of new types of network application scenarios, and the traditional Internet is difficult to meet the demands of different types of business services. Traditional routing algorithms usually make decisions based on network topology, and changes in network topology and scenarios lead to frequent updates of routing tables and network congestion. In order to solve the problem of reducing the dependence on the network topology and improving the fault tolerance, adaptivity and scalability of the network under the demand of multi-service type of network services, this paper proposes the TFIR algorithm, which employs segmented routing to dynamically select the best path for packet forwarding, making the algorithm better adapt to the changes of the new topology. TFIR is a GCN-driven Distributed Multi-Intelligent Reinforcement Learning Algorithm, which uses distributed intelligent reinforcement learning to train and obtain generalized SR intelligences to realize topo-free routing algorithms for multi-service type requirements. The algorithm uses GCN network to capture the hidden information in the network topology state and express it using a generic form. Experiments show that the algorithm outperforms some widely used traditional routing algorithms and has higher scalability and adaptive capabilities.