Network anti-inference: A fundamental perspective on proactive strategies to counter flow inference

Network inference is an effective mechanism to infer end-to-end flow rates and has enabled a variety of applications (e.g., network surveillance and diagnosis). The paper is focused on the opposite side of network inference, i.e., how to make inference inaccurate, which we call network anti-inference. As most research efforts have been focused on developing efficient inference methods, design of anti-inference is largely overlooked. Anti-inference scenarios can rise when network inference is not desirable, such as in clandestine communication and military applications. Our objective is to explore network dynamics to provide anti-inference. In particular, we consider two proactive strategies that cause network dynamics: transmitting deception traffic and changing routing to mislead the inference. We build an analytical framework to quantify the induced inference errors of the proactive strategies that maintain limited costs. We find via analysis and simulations that for deception traffic, a simple random transmission strategy can achieve inference errors on the same order of the best coordinated transmission strategy; while changing routing can cause inference errors of higher order than any deception traffic strategy. Our results not only reveal the fundamental perspective on proactive strategies, but also offer the guidance into practical design of anti-inference.