A Microeconomic Approach to Intelligent Resource Sharing in Multiagent Systems

We have analyzed characteristics of sharable resources and developed techniques for intelligently sharing resources--specifically, communication channels-among agents in multiagent systems. Our techniques allow agents to nearly optimize their communication behavior in a distributed fashion, involving the use of a microeconomic pricing system based on economic laws of supply and demand and trading among agents in real-time. Our ~alyses are based on three measures of performance: fairness of resource allocation, utilization of resources, and waiting time for resources. Our initial analysis indicates that fairness and utilization conflict under any statically-defined resource allocation policy. Suppose that two agents, a and b attempt to send an infinite number of messages through the shared communication channel with probability, called transmission-potential, P, and Pb, respectively. The utilization of the channel is P.(1-Pb) +Pb(1--P~). On the one hand, with a fair policy (P~ Pb), the mazimum utilization, .5, occurs at P~ -Pb -.5. On the other hand, with a biased policy (Pa -1 -Pb), the minimum utilization, .5, occurs at Pa = 1 Pb -.5. From this analysis, we conclude that when there are infinite messages to send, the best "fair" allocation of potentials corresponds to the worst "unfair" allocation. Utilization, however, is not the only interesting parameter. The mean message delay is also important for timely communication. Unfortunately, high utilization and low delay are inherently in conflict. For example, if one agent alone has all the potential, the channel utilization is maximized to 1 (assuming iv6nite pending messages), but the mean message delay is infinite. The simplest way to overcome this problem is to alternate transmission potentials between the two agents, so they take turns with the cllannel. Simply swapping 1/0 potentials works very well with infinite messages, but if we relax this assumption, some bandwidth of the channel may be wasted because the agent may have no message to send while