Component and data distribution in a distributed workflow

Designing a distributed workflow management system (WFMS) has become a topic with tremendous research interests in recent years. Unlike a traditional client-server WFMS that executes only at a single site, a distributed WFMS allows the execution of a workflow instance to be governed by more than one servers so as to achieve availability and efficiency. A consensus has generally been reached that the component-based approach is a promising one to designing a distributed WFMS. The component-based approach divides the functionality of a WFMS into a number of components, each of which follows some well-accepted distributed object standard, e.g. CORBA or DCOM. These components can be distributed transparently to different sites for execution, and cooperatively they achieve the goal of the underlying business process. Following the component-based approach, this paper proposes and discusses several alternatives on how to distribute the components and the data of a distributed WFMS. Specifically, the component distribution problem can be transformed into two combinatorial problems, namely the integer programming problem and the weighted matching problem of a bipartite graph under different conditions. Regarding the data distribution, we recommend the full replication of workflow definition data and suggest different strategies in placing different types of instance data.