Robustness of Cognitive Networking Mechanisms to Malicious Insiders.

摘要

Cognitive networking mechanisms promise to greatly improve network performance over non-cognitive mechanisms, by making more efficient use of bandwidth, spectrum, and power. However, these mechanisms must be designed with cyber security in mind in order to remain efficient in the presence of subverted, adversarial participants. In this paper, we demonstrate the susceptibility of two specific cognitive-networking mechanisms to a single Byzantine participant. Specifically, we describe a novel 'energy well' attack against Q-routing, in which a Byzantine participant can attract traffic meant for an honest participant. Secondly, we describe a denial of service attack against a no-regret learning algorithm for Dynamic Spectrum Access (DSA), in which a single Byzantine participant can degrade network-wide performance for an arbitrary amount of time. These attacks demonstrate why cyber security techniques must be designed into cognitive mechanisms before use in the tactical field so that they do not fail to tolerate adversarial behavior. We conclude by discussing possible mitigation concepts and future work.