- Beta Distribution

Abstract

Current service-oriented architectures lack business-level software abstractions. As a result, existing implementations are unnecessarily rigid. This dissertation takes as its point of departure the idea that agents who offer and consumer business services enter into commitments with one another. These commitments give meaning to the interaction among the agents. Commitments support a semantic notion of compliance and enable flexible enactment of business processes.Interoperability refers to the ability of agents to engage in interaction with oneanother. In open systems, where agents are autonomous and heterogeneous, ensuring interoperability is critical. Traditionally, interoperability has been formulated in low-level terms. This dissertation presents commitment alignment as a key form of business-level interoperability. Agents are aligned if whenever the creditor of a commitment infers the commitment, the debtor infers it too. A misalignment precludes any possibility of successful engagement among agents-their interaction would break down.This dissertation formally characterizes alignment in multiagent settings. It presents the causes of misalignment, namely, autonomy, distribution, and heterogeneity. Autonomy means that agents communicate asynchronously; distribution refers to the fact that in distributed systems, some agents may have more information than others; and heterogeneity refers to the fact that agents may have incompatible interfaces.To address autonomy and distribution, we propose a formalization of commitments that consists of three elements: a semantics of the commitment operations; messaging patterns that implement the commitment operations; and weak constraints on agents' behaviors to ensure the propagation of vital information. The constraints result in messages that are critical to alignment. We prove that under our formalization, no misalignment occurs because of autonomy or distribution. To address heterogeneity, we propose a language for agent interfaces, formulate a decision procedure that checks for interface compatibility, and prove its correctness. By addressing all three causes, we guarantee that no misalignment occurs.