Recognizing BGP Communities Based on Graph Neural Network

The identification of BGP community attributes holds significant importance in routing modeling for understanding their impact on routing behavior and policies. However, when it comes to identifying BGP community attributes, existing rule-based solutions suffer from low automation and coarse granularity for relying on human knowledge, and the immutability of rule-based solutions makes them impractical for time-sensitive applications. Consequently, our research proposes a consistent and automated approach for identifying BGP community attributes in AS-level. We pick out the related autonomous system (AS) paths and community attribute numeric identifiers within BGP forwarding entries. Using these AS paths and identifiers we construct a topological graph structure and generate relevant feature embeddings for nodes and edges. Subsequently, we build a graph neural network (GNN) model consisting of a residual network for convolution and a fully connected layer, which can preserve and highlight the differences in the features of different communities. We employ this model to classify BGP communities. It turns out that the accuracy of our design can surpass 96%. According to the experimental result, our approach outperforms other state-of-the-art methods.