Self-Adaptive Dissemination of Critical Data in Dynamic Wireless Networks

Abstract : The distribution of data in large-scale dynamic wireless networks presents a difficult problem for network designers due to node mobility, unreliable links, and traffic congestion. In this work, we propose a framework for adaptive data dissemination protocols suitable for distributing data in large-scale dynamic networks without a central controlling entity. The framework consists of cooperating mobile agents and a reinforcement-learning component with value function approximation. A component for agent coordination is provided, as well as rules for agent replication, mutation, and annihilation. We examine the adaptability of this framework to a data dissemination problem in a simulation experiment and discuss potential benefits of this framework for the Future Combat System (FCS) Network. The large-scale, dynamic, and time-varying nature of operational environments for ad hoc wireless mobile and sensor networks present formidable challenges to the design of reliable dissemination protocols. Approaches based on centralized control over the network are often infeasible because they do not scale well and assume, unrealistically, a static structure in the form of routes or routing tree structures. Decentralized approaches that do not assume any structure in the network often rely upon a gradient that emerges as data flow toward a sink node. The problem with these structure-less approaches is that they assume the gradient is relatively static; all broadcasts originate from a small subset of nodes and all data flow to these same nodes. However, if the network is deployed for the purpose of communicating with mobile or static nodes working within the area of deployment, as is the case with networks envisioned for use by Brigade Combat Teams (BCTs) in the Future Combat System (FCS), these traditional protocols are no longer suitable because any node may be the source of a broadcast to all other nodes.