Defending against fake VIP in scant-transparency information systems with QoS differentiation

Abstract

In client–server information systems with quality of service (QoS) differentiation, Client may deplete Server’s resources by demanding unduly high QoS level. Such QoS abuse has eluded systematic treatment; known defenses using Client authorization, payments, or service request inspection prior to QoS assignment, are heuristic and environment-specific. We offer a game-theoretic approach on the premise that a service request is occasionally trusted to reduce the inspection cost. We call Fake VIP attack (FVA) a form of QoS abuse that consciously exploits Server’s trust. An FVA strategy instills trust to maximize Client’s utility gained from successful FVAs, whereas a trust strategy maximizes Server’s utility by trading her loss due to successful FVAs against the request inspection cost. We consider a realistic scant-transparency setting where only long-term utilities are observable. Against a probabilistic FVA strategy we design a trust strategy based on double-blind reputation. Assuming a memoryless service request stream we analyze the impact of the request inspection cost and information leakage on the utilities at the Stackelberg equilibrium of the arising game. Experimental comparison with a real-world internally correlated stream is also shown.