Distributed Trust Management in Wireless Autonomic Networks

As an important concept in network security, trust is interpreted as a set of relations among agents participating in the network activities. Trust relations are based on previous behaviors of agents as well as on trust documents. We present our results on distributed trust management in MANET. The trust information or evidence used to evaluate trustworthiness is provided by peers, i.e. the agents that form the network. We describe first a new trust document distribution scheme based on swarm intelligence. Then we describe various methods for distributed trust evaluation and the associated trust (and mistrust) spreading dynamics. Under such dynamics the whole network evolves as the local interactions iterate from isolated trust islands to a connected trust graph. Our interest is to discover rules and policies that establish trust-connected networks using only local interactions, to find the conditions under which trust spreads to a maximum set, as well as the parameters (e.g. topology type) that speed up or slow down this transition. We analyze the dynamics induced by local interaction rules using algebraic graph theory and methods inspired from the statistical physics of spin-glass materials. We describe and explain the phase transition phenomena that we have found in these evolutions. We model the interactions among agents as cooperative games and show that trust can encourage agents to collaborate. We also describe a model for trust evaluation that uses pairwise iterated graph games between the agents to create a trust reputation with evolution coupled to the game dynamics. Finally we present a new modeling framework for trust metric evaluation as linear iterations over ordered semirings, by treating such evaluations as path problems in MANET. This allows us to formulate problems of resilience of trust metrics and trust evaluation to attacks.