New strategies for revocation in ad-hoc networks

Responding to misbehavior in ad-hoc and sensor networks is difficult. We propose new techniques for deciding when to remove nodes in a decentralized manner. Rather than blackballing nodes that misbehave, a more efficient approach turns out to be reelection - requiring nodes to secure a majority or plurality of approval from their neighbors at regular intervals. This can be implemented in a standard model of voting in which the nodes form a club, or in a lightweight scheme where each node periodically broadcasts a 'buddy list' of neighbors it trusts. This allows much greater flexibility of trust strategies than a predetermined voting mechanism. We then consider an even more radical strategy still - suicide attacks - in which a node on perceiving another node to be misbehaving simply declares both of them to be dead. Other nodes thereafter ignore them both. Suicide attacks, found in a number of contexts in nature from bees to helper T-cells, turn out to be more efficient still for an interesting range of system parameters.