Résumé: In trust negotiation and other forms of distributed proving, networked entities cooperate to form proofs of authorization that are justified by collections of certified attribute credentials. These attributes may be obtained through interactions with any number of external entities and are collected and validated over an extended period of time. Although these collections of credentials in some ways resemble partial system snapshots, current trust negotiation and distributed proving systems lack the notion of a consistent global state in which the satisfaction of authorization policies should be checked. In this article, we argue that unlike the notions of consistency studied in other areas of distributed computing, the level of consistency required during policy evaluation is predicated solely upon the security requirements of the policy evaluator. As such, there is little incentive for entities to participate in complicated consistency preservation schemes like those used in distributed computing, distributed databases, and distributed shared memory. We go on to show that the most intuitive notion of consistency fails to provide basic safety guarantees under certain circumstances and then propose several more refined notions of consistency that provide stronger safety guarantees. We provide algorithms that allow each of these refined notions of consistency to be attained in practice with minimal overheads and formally prove several security and privacy properties of these algorithms. Lastly, we explore the notion of strategic design trade-offs in the consistency enforcement algorithm space and propose several modifica- This research was supported by the NSF under grants IIS-0331707, CNS-0325951, and CNS- 0524695 and by Sandia National Laboratories under grant number DOE SNL 541065. A.J. Lee was also supported in part by a Motorola Center for Communications graduate fellowship.

Langue: Anglais