Poaching and distraction in asynchronous agent activities

We investigate coordination issues in a distributed jobshop scheduling system in which agents schedule potentially contentious activities asynchronously in parallel. Agents in such a system will in general have a limited view of the global state of resources and must exchange appropriate state information with other agents in order to schedule effectively. However, even given perfect instantaneous knowledge of other agents' resource requirements, agents still may not be able to schedule effectively if they do not also model the possible future actions of other agents and the effects of their own actions. We formally describe two types of agent behaviors, poaching and distraction, arising from the asynchronous nature of distributed systems that decrease scheduling effectiveness, and we present experimental results from a distributed airport resource management system demonstrating a significant improvement in scheduling performance when coordination mechanisms are used to prevent such behaviors.