Hierarchical Dependence Graphs for Dynamic JDF Workflows

Abstract

The Hierarchical Dependence Graph (HDG) described in this paper extends the theory of directed acyclic graph (DAG) [1,21 by allowing hierarchical representation of workflows. It can be used to explicitly express the dependences across JDF [31 (process) nodes and resources derived from any JDF job description. It defines a flexible and semantic-rich model to represent JDF workflow as a set of DAGs at different abstractions: intent level, process group levels and process execution level. By explicitly representing JDF workflows in the HDG, not only it enables the separation of the workflow itself from MIS/or Controller implementations to support fully dynamic JDF workflows, but also it provides a theoretic basis for formal analysis of JDF workflows. Furthermore, we introduce the concept of Connectivity Matrix and its transformations to allow two views derived from a single model: process-centric view and resource-centric view. By exploiting the fact that each of these views is a DAG with a hierarchical structure, we then show how various analytical properties defined for DAG can be recursively applied to analyze JDF workflows, particularly in the following perspectives: (1) validating JDF workflows in terms of cyclic dependence, missing resources and dangling resources; (2) identifying the impacted JDF nodes or resources due to the resource availability and workflow status changes; (3) intelligently handling failures or exceptions by considering causal relations between resources and processes.