Coordination to Avoid Starvation of Bottleneck Agents in a Large Network System

Rajesh Gautam, Kazuo Miyashita

In this paper, we present a multi-agent control method for a large-scale network system. We propose an extension of a token-based coordination technique to improve the tradeoff between two conflicting objectives of the network system: reducing the lead time and increasing throughput. In our system, CABS, information about an agent's urgency of jobs to fulfill demanded throughput and to maintain its utilization is passed from downstream agents in the network so that upstream agents can provide necessary and sufficient jobs to bottleneck agents whose loss of capacity degrades the total system performance. We empirically evaluate CABS performance using a benchmark problem of the semiconductor fabrication process, which is a good example of a large-scale network system.