A Framework for Optimizing Distributed Workflow Executions

A central problem in workflow concerns optimizing the distribution of work in a workflow: how should the execution of tasks and the management of tasks be distributed across multiple processing nodes (i.e., computers). In some cases task management or execution may be at a processing node with limited functionality, and so it is useful to optimize translations of (sub-)workflow schemas into flowcharts, that can be executed in a restricted environment, e.g., in a scripting language or using a flowchart-based workflow engine. This paper presents a framework for optimizing the physical distribution of workflow schemas, and the mapping of sub-workflow schemas into flowcharts. We provide a general model for representing essentially any distribution of a workflow schema, and for representing a broad variety of execution strategies. The model is based on families of “communicating flowcharts” (CFs). In the framework, a workflow schema is first rewritten as a family of CFs that are essentially atomic and execute in parallel. The CFs can be grouped into “clusters”. Several CFs can be combined to form a single CF, which is useful when executing a sub-schema on a limited processor. Local rewriting rules are used to specify equivalencepreserving transformations. We developed a set of formulas to quantify the metrics used for choosing a near optimal set of CF clusters for executing a workflow. The current paper focuses primarily on ECA-based workflow models, such as Flowmark, Meteor and Mentor, and condition-action based workflow models, such as ThinkSheet and Vortex.