Removing the blinders: Using information to mitigate adversaries in adaptive overlays

The proliferation of peer-to-peer systems has led to the increasing deployment of dynamic, adaptive overlay networks that are designed to preserve application performance goals. While such networks provide increased performance and resiliency to benign faults, they are susceptible to attacks conducted by compromised overlay nodes, especially those targeting the adaptation mechanisms. In this work, we propose a lightweight, general solution to increase the resiliency of adaptive overlay networks. By locally aggregating and correlating network topology with system performance metrics such as latency and bandwidth, each node can check the consistency of the reported information and constrain the attacker's ability to lie about system metrics. As a result, each node can make better adaptation decisions. We demonstrate the susceptibility of adaptation mechanisms to malicious attacks and the utility of our solution through real-life deployments of mature, adaptive overlay-based systems.