An Agent Based Approach for the Decentralised Compensation of Business Processes

Abstract

An automated execution of business processes, composed of Web Services, also requires a reliable error handling and in the case of failures at least parts of a process need to be recovered. One way to enable a forward oriented recovery is compensation which allows to semantically undo effects even "long" time after the process --a long running transaction-- commits. Usually, compensation as for example applied in BPEL (Business Process Execution Language), requires that all compensation steps associated with the business process perform successfully. Also, the conditions for a compensation are not considered and it is not dynamic. In some business scenarios, e.g. a complex production process involving several partners, compensation is a commitment under certain conditions. These conditions in turn can be dynamic and interdependent. Often, a lot of human effort is required to perform compensation. Agents in turn, can enable a decentralised compensation in which agents negotiate a commitment to find a proper compensation strategy. Especially in highly dynamic environment this enables on the one hand a dynamic compensation, and on the other it can support human decision finding in case of failure.