The Value of Fairness: Trade-offs in Repeated Dynamic Resource Allocation

Resource allocation problems are an important part of many distributed autonomous systems. In sensor networks, they determine which nodes get to use the communication links, in SmartGrid applications they decree which electric vehicle batteries are loaded, and in autonomous power management they select which generators produce the power required to satisfy the overall load. These cases have been considered in the literature before under the aspect of demand satisfaction: how well can distributed algorithms with local knowledge approximate the best allocation. A factor that has been ignored, however, is fairness: how fair is the resource allocation and -- in extension -- the distribution of revenue, wear, or recovery time. In this paper, we bring together previously disjoint approaches on dynamic distributed resource allocation and on fairness in electronic institutions. We show that fair allocations based on Ostrom's principles and on Rescher's canons of distributive justice create value in repeated resource allocations. We apply the scheme to solve the multi-objective problem of distributing load to generators fairly based on demands made by the individual generators. Our evaluation shows that a fair distribution increases satisfaction of the individual agents while reducing the hazard of optimising the problem in the short-term at the cost of long-term robustness and stability.