Our work may be most similar to the use of session directories for Internet multicast address allocation described by Handley [8]. Multicast addresses are selected randomly in a decentralized manner by nodes that need to establish multicast groups. The idea parallels AFF's use of temporal locality (multicast addresses are reused over time) and spatial locality (TTL scoping prevents multicast groups from being globally visible). In contrast, our focus on sensor networks brings with it different constraints such as the paramount importance of energy efficiency.
In the Nimrod architecture [3], Chiappa recognizes problems that arise when addresses serve multiple, overloaded roles. He suggests an architecture in which nodes have a globally unique endpoint identifier separate from their unique and topologically significant ``locator.'' This allows endpoints to have a stable identity regardless of changes in the network topology. In contrast, our work is more concerned with the cost of identifiers long enough to be globally unique in context where locally unique identifiers are sufficient. Our scheme also assumes some other method is used for naming, and names need not be unique.
In WINS [2], Kaiser and Pottie have designed a system where short, locally unique addresses are dynamically assigned to nodes in a radio cluster by a central controller. They try to maximize the use of energy in a wireless, unattended radio system by reducing the number of address bits transmitted. Their motivation is similar to AFF; however, AFF's design does not require centralized cluster formation. This makes AFF more scalable, feasible without a centralized controller, and robust in the face of high dynamics.
In the MIT Amorphous Computing project, Coore et. al. have described algorithms for hierarchy construction that rely on randomized node identifiers to break ties in elections [4]. Our work, in contrast, is oriented towards identifiers used to facilitate transactions, and does not deal with cluster formation.
In SCADDS [6], EGHK propose an architecture for scalable coordination and control in deeply distributed systems such as sensor networks. Our work is complementary to SCADDS and we owe many of our assumptions about future sensor network architectures to that project. The attribute-based data naming proposed in SCADDS is similar to the naming schemes used by Raman and McCanne in ALF [17,7], Adjie-Winoto et. al. in the Intentional Naming System [1], and Michel et. al. in their adaptive web caching architecture [13].