Centralized Traffic Engineering (TE) plays a critical role in network management, owing to its potential to achieve optimal or near-optimal network performance. However, the overhead from frequent network-wide traffic measurements required to observe the global network view significantly limits its practical use. To address this issue, we propose an end-to-end learning approach called TEST for routing optimization using sparse Traffic Matrices (TMs) obtained from sparse traffic measurements, which require measuring traffic on only a subset of network nodes. Specifically, to mitigate the impact of unknown traffic demands in unmeasured nodes on network performance, we construct a set of synthesized TMs with diverse traffic patterns adaptively to enhance the robustness of the generated routing policies. To address the missed information in sparsely measured TMs, we leverage historical sparse TM sequences to provide sufficient evidence for generating routing policies. To effectively capture the spatio-temporal relationships in the sparse TM sequence, we propose designing a routing model by integrating a Transformer model with a Graph Convolutional Network (GCN). Extensive experiments conducted on topologies with different scales demonstrate that the proposed TEST achieves promising TE performance under sparse traffic measurements. Additionally, discussions on scenarios involving network failures and traffic changes further highlight its robustness.

Routing optimization, as a significant part of Traffic Engineering (TE), plays an important role in balancing network traffic and improving quality of service . With the application of Machine Learning (ML) in various fields, many neural network-based routing optimization solutions have been proposed. However, most existing ML-based methods need to retrain the model when confronted with a network unseen during training, which incurs significant time overhead and response delay. To improve the generalization ability of the routing model, in this paper, we innovatively propose a routing optimization method GROM which combines Deep Reinforcement Learning (DRL) and Graph Neural Networks (GNN), to directly generate routing policies under different and unseen network topologies without retraining. Specifically, for handling different network topologies , we transform the traffic-splitting ratio into element-level output of GNN model . To make the DRL agent easier to converge and well generalize to unseen topologies, we discretize the huge continuous traffic-splitting action space. Extensive simulation results on five real-world network topologies demonstrate that GROM can rapidly generate routing policies under different network topologies and has superior generalization ability.