Feedback Verification for Trustworthy Tomography

Network tomography is a process by which internal characteristics of a network are inferred from “external” end-to-end measurements. To ensure that the inferred internal characteristics are sound and trustworthy, it is essential to verify the integrity of data collected from external measurements. In this paper, we present an algorithm which can verify the integrity of data collected from end-to-end multicast measurements. This data is used by a multicastbased tomography tool called MINC to infer loss rates on internal network links. MINC performs loss inference by analyzing binary feedbacks reported by receivers in response to multicast probes sent from the source. However, buggy or malicious receivers can report incorrect feedbacks, resulting in a faulty loss inference. In this work, we consider the problem of verifying the integrity of binary feedbacks collected from receivers of a multicast tree to ensure a sound and trustworthy MINC loss inference. We start by showing how the MINC loss inference becomes erroneous if feedbacks of receivers are altered. Then, we present a statistical verification algorithm which checks if feedbacks of receivers are consistent with respect to one another. We present the performance of this algorithm on Model-based traces, NS traces and MBone loss traces.