Title:Federated Route Leak Detection in Inter-domain Routing with Privacy Guarantee

Abstract:In the inter-domain network, a route leak occurs when a routing announcement is propagated outside of its intended scope, which is a violation of the agreed routing policy. The route leaks can disrupt the internet traffic and cause large outages. The accurately detection of route leaks requires the share of AS business relationship information of ASes. However, the business relationship information between ASes is confidential due to economic issues. Thus, ASes are usually unwilling to revealing this information to the other ASes, especially their competitors. Recent advancements in federated learning make it possible to share data while maintaining privacy. Motivated by this, in this paper we study the route leak problem by considering the privacy of business relationships between ASes, and propose a method for route leak detection with privacy guarantee by using blockchain-based federated learning framework, in which ASes can train a global detection model without revealing their business relationships directly. Moreover, the proposed method provides a self-validation scheme by labeling AS triples with local routing policies, which mitigates route leaks' lack of ground truth. We evaluate the proposed method under a variety of datasets including unbalanced and balanced datasets. The different deployment strategies of the proposed method under different topologies are also examined. The results show that the proposed method has a better performance in detecting route leaks than a single AS detection regardless of whether using balanced or unbalanced datasets. In the analysis of the deployment, the results show that ASes with more peers have more possible route leaks and can contribute more on the detection of route leaks with the proposed method.