Distributed Resource Allocation with Side Payments

Distributed resource allocation problems (DRAPs) are composed of multiple agents and of resources that need to be allocated among them, and can be easily represented as asymmetric distributed constraints optimization problems (ADCOPs). Standard methods for solving ADCOPs use distributed multi-agent search to find solutions of high global social welfare. The present paper focuses on a family of resource allocation problems - allocating operating rooms to medical wards in a large hospital - that also benefit from allocations with increased fairness. The proposed multi agent algorithm has agents that represent ward managers that use a partially cooperative ADCOP algorithm. The proposed DRAP algorithm uses a new approach to partial cooperation - agents can offer or propose (non-strategic) side payments in exchange for resources. The use of side payments enables the agents to perform transactions that increase the global social welfare while retaining their own utility throughout the search. The proposed method is experimentally compared to an existing model of partial cooperation in a recent study of these problems. It is found to improve all measured objectives - both the utilitarian social welfare and well known fairness measures.