LSA: Understanding the Threat of Link-based Scapegoating Attack in Network Tomography
Xiaojia Xu, Yongcai Wang, Lanling Xu, Deying Li · IEEE Transactions on Network Science and Engineering · 2023
Network tomography is a powerful and convenient tool to infer the internal states of a network via end-to-end path measurements. However, this paper shows that when identifiability is not satisfied at all links (which is general in applications for cost saving purpose), a special kind of attack, i.e.,link-based scapegoating attack (LSA)that can greatly degrade the network performance while misleading the network tomography to identify some normal links as problematic links can be launched. To fully understand the threat from LSA, this paper in particular investigates the conditions and efficient methods to launch LSA from an attacker's point of view. Specifically, an efficient algorithm to figure out the potential scapegoat links is first proposed, and the conditions to launch LSA are presented. Then aminimum-cost scapegoating attack (MCSA)problem is proposed, which studies how to manipulate the least-cost link sets to launch LSA. A weighted set cover model and a greedy approximation algorithm are designed to solveMCSAproblem, which is approached by a linear programming method with$H_{K}$approximation ratio. Evaluation on both synthetic and real network topologies shows the wide existence of LSA threats, the feasibility of the proposed LSA conditions, and the attacking strategies.