Distributed Sub-gradient Method for Delay Tolerant Networks
Riccardo Masiero, Giovanni Neglia · HAL (Le Centre pour la Communication Scientifique Directe) · 2010
In this paper we consider that nodes in a Delay Tolerant Network (DTN) may collaborate to minimize the sum of local objective functions, depending in general on some parameters or actions of all the nodes in the network. If the local objective functions are convex, it can be adopted a recently proposed computation framework, that relies on local sub-gradient methods and consensus algorithms to average each node information. Existing convergence results for this framework can be applied to DTNs only in the case of nodes' synchronous operation and simple mobility models without memory. We address both these issues. First, we prove convergence to the optimal solution for a more general class of mobility processes. Second, we show that, under asynchronous operation, a straight application of the original method would lead to suboptimal solutions and we propose some changes to solve this problem. As a particular case study, we show how the framework can be applied to optimize the dissemination of dynamic content in a DTN.