Toward Greater Intelligence in Route Planning: A Graph-Aware Deep Learning Approach

Software-defined networking decouples the control plane and data plane, which grants more computing power for routing computations. Traditional routing methods suffer from the complex dynamics in networking, and they are facing issues such as slow convergence and performance decline. Deep learning techniques have shown preliminary results on solving the routing problem, and bringing more accuracy, precision, and intelligence compared with traditional modeling techniques. However, the existing deep learning architectures are not built to learn from the crucial topological relations between forwarding nodes, which restricts the model's ability to handle different network conditions. In this paper, we propose a deep learning based intelligent routing strategy with revised graph-aware neural networks, which learns topological information efficiently. In addition, a set of features suitable for network routing are designed so that the networking state are well represented upon each routing decision. In experiments, the performance of the proposed work is demonstrated with a real-world topology and the production level software switches. The execution time is evaluated on various kinds of network topology and different network scales. Also, the simulation result shows that the proposed work is more accurate and efficient compared to the state-of-the-art routing strategy.