Using uncertainty in reputation methods to enforce cooperation in ad-hoc networks

Proceedings of the 5th ACM workshop on Wireless security |

Published by Association for Computing Machinery, Inc.

http://doi.acm.org/10.1145/1161289.1161308

This paper gives an approach to reputation computation that incorporates “uncertainty” based on subjective logic. Uncertainty can arise when a node joins the network and thus has no history, or when a node’s behavior changes in a way that is not clearly malicious, but at least suspicious. This uncertainty indicates when a node evaluating the reputation rating of another node should poll its neighbors for recommendations, because its local opinion is not sufficiently well-informed.The uncertainty based algorithm has several parameters and the computed reputation depends on these parameters. To evaluate the effectiveness of the uncertainty based method and to determine the effect of the parameters on the reputation computations, we have implemented a simulation infrastructure for the behavior of ad-hoc networks using the uncertainty based scheme. The simulator can also use other previously proposed mechanisms for reputation computation. The simulator is used to determine optimal parameters for the various schemes and how well each scheme a) ensures cooperative nodes receive service when requested, and b) throttles the ability of free-loading (malicious) nodes to consume service without contributing. The simulations show that the reputation computation that incorporates uncertainty provides superior recognition of “malicious” behavior in a variety of ad-hoc network situations. These initial results are sufficiently promising to motivate continuing study of reputation computations incorporating uncertainty to additional models of malicious behavior and a broader range of network behaviors.