Robust Decentralized Virtual Coordinate Systems in Adversarial Environments

Abstract

Virtual coordinate systems provide an accurate and efficient service that allows hosts on the Internet to determine the latency to arbitrary hosts without actively monitoring all of the nodes in the network. Many of the proposed systems were designed with the assumption that all of the nodes are altruistic. However, this assumption may be violated by compromised nodes acting maliciously to degrade the accuracy of the coordinate system. As numerous peer-to-peer applications come to rely on virtual coordinate systems to achieve good performance, it is critical to address the security of such systems. In this work, we demonstrate the vulnerability of decentralized virtual coordinate systems to insider (or Byzantine) attacks. We propose techniques to make the coordinate assignment robust to malicious attackers without increasing the communication cost. We use both spatial and temporal correlations to perform context-sensitive outlier analysis to reject malicious updates and prevent unnecessary and erroneous adaptations. We demonstrate the attacks and mitigation techniques in the context of a well-known virtual coordinate system using simulations based on three representative, real-life Internet topologies of hosts and corresponding Round Trip Times (RTT). We show the effects of the attacks and the utility of the mitigation techniques on the virtual coordinate system as seen by higher-level applications, elucidating the utility of deploying robust virtual coordinate systems as network services.