The value of side information in network flow optimization

Consider a group of agents who seek to simultaneously traverse a graph. Each edge of the graph has an associated weight (e.g., a delay), and the agents seek to minimize the cumulative weight incurred by all agents as each traverses a path of the graph. An edge’s weight is a function of the number of agents that use that edge as well as an inherent random weight. If the agents have no side information about the random component, they will (deterministically) organize themselves so as to optimize their average performance. We consider a generalization of this framework whereby the agents have access to a limited amount of shared side information about the edge weights, and we study the relationship between information quantity and performance.