Path Planning Method using Deep Learning Model for Traffic Engineering in Small Networks

Traffic engineering is a method for efficiently con-trolling traffic with large time fluctuations: while maintaining the Quality of Service (QoS) level for the user. On the other hand, it is hard to realize traffic engineering in a realistic network. This technology has problems, for example, difficulty recalculating the link cost according to the network conditions, deterioration of scalability according to the network scale. To solve this problem, we try to solve it using machine learning. In particular, we propose a path planning method using the deep learning model for traffic engineering. In the conventional path design method using machine learning, each router has a model in a distributed manner and designs the next hop. That is to maintain the path estimation accuracy at a high value. However, in traffic engineering, it is necessary to design the path considering the entire network, and it is hard to learn the features with the conventional model. In this paper, we define the deep learning model and the teacher signal so that the entire path from source to destination nodes can design. The experimental result shows that the entire path is designed with very high accuracy and that the proposed method is superior in terms of calculation time while achieving load balancing comparable to that of Dijkstra’s algorithm.