Autonomous agents in service-oriented negotiation: Strategy, protocol, and coordination
Due to the convergence of industrial demands for business-process and supply-chain management and recent results in multiagent systems, autonomous software services, and the Semantic Web, Web services are becoming the main focus for the next generation of the Internet. They will behave like intelligent agents by composing themselves cooperatively into workflows. A workflow is a set of services that execute by carrying out specified control and data flows. Agents are persistent active entities that can perceive, reason, and act in their environment, and communicate with other agents. Agents can interact autonomously across enterprise boundaries and, when thought of as services, provide a new way to achieve programming-in-the-large. Agents interact with other agents through negotiation since they are autonomous entities. Negotiation is a technique for reaching mutually beneficial agreement through com¬munication among agents. Agents can autonomously negotiate and coordinate with others to execute a workflow in a heterogeneous system by agreeing on functional and quality metrics of the services they request and provide. The focus of this dissertation is on negotiating and coordinating a composed service represented as a workflow among self-interested service agents in a competitive service-oriented environment. By balancing between optimized utility and computation cost, I introduce an optimal strategy for multiple-issue negotiation between interacting agents with bounded rationality. As the number of services available on the Web proliferates, it is very likely that multiple service requestors and providers will be negotiating simultaneously and competitively. I present a protocol to support a many-to-many, bilateral, multiple-issue negotiation. This protocol is neutral to both service requestors and service providers. Therefore, it can eliminate the decommitment situations arising in one-sided commitment. Commitments among agents can be used to model a work-flow and coordinate their execution of it. I provide methodologies to map an OWL-S model for a workflow to a Colored Petri Net, a graphical and mathematical modeling tool for information processing systems, and then infer commitments and causal relationships from the CPN graph. This methodology enables agents to enact a workflow collaboratively through commitment-based formalisms.