A workflow can be deemed as a collection of cooperating and coordinated activities designed to carry out a well-defined complex process, such as a trip planning, conference registration procedure, or business process in an enterprise. A workflow model is used to describe a workflow in terms of various elements, such as roles and resources, tools and applications, activities, and data, which represent different perspectives of a workflow [15, 16]. Roles and resources elements represent organizational perspective that describes where and by whom tasks are performed and available resources tasks can utilize in the organization. Tools and applications elements represent operational perspectives by specifying what tools and applications are used to execute a particular task. Activity elements are defined with two perspectives: 1) functional: what tasks a workflow performs; and 2) behavioral: when and how tasks are performed. Data elements represent the informational perspective, i.e., what information entities are produced or manipulated in the corresponding activities in a workflow.
A well-defined workflow model leads to the efficient development of an effective and reliable workflow application. The correctness issues in a workflow might be classified into three dimensions: control-flow, resource, and data-flow. Generally, the analyses in control-flow dimension are focused on correctness issues of control structure in a workflow. The common control-flow anomalies include deadlock, livelock (infinite loop), lack of synchronization, and dangling reference [17－28]. A deadlock anomaly occurs if it is no longer possible to make any progress for a workflow instance, e.g. synchronization on two mutually exclusive alternative paths. A livelock anomaly indicates an infinite loop, such as iteration without possible exit condition, which causes a workflow to make continuous progress, however, without progressing toward successful completion. A lack of synchronization anomaly represents the case of more than one incoming vertex merging into an or-join vertex. Activities without termination or without activation are two common cases of dangling reference anomaly.
Activities belonging to different workflows or parallel activities in the same workflow might access the same resources. A resource conflict occurs when these activities execute over the same time interval. Thus, the analyses in resource dimension include the identification of resource
conflicts under resource allocation constraints and/or under the temporal and/or causality constraints [2－6]. On the other hand, missing, redundancy, and conflict use of data are common anomalies in data-flow dimension [7－10]. A missing data anomaly occurs when an artifact is accessed before it is initialized. A redundant data anomaly occurs when an activity produces an intermediate data output but this data is not required by any succeeding activity. A conflicting data anomaly represents the existence of different versions of the same artifact.
Current workflow modeling and analyzing paradigms are mainly focused on the soundness of control logic, i.e., in the control-flow dimension, including process model analysis [19－30], workflow patterns [20－33] and automatic control of workflow process [34]. Aalst and ter Hofstede [19] proposed a WorkFlow net (WF-net), based on Petri nets, to model a workflow:
transitions representing activities, places representing conditions, tokens representing cases, and directed arcs connecting transitions and places. Furthermore, control-flow anomalies, such as deadlock, livelock, and dangling reference (activities without termination or activation) have been identified through Petri net modeling and analysis. Son [35] defined a well-formed workflow based on the concepts of closure and control block. He claimed that a well-formed workflow is free from structural errors, and that complex control flows can be made with nested control blocks. Son [35] and Chang [36] identified and extracted the workflow critical path from the context of the workflow schema. They proposed extraction procedures from various non-sequential control structures to sequential paths, thus obtaining appropriate sub-critical paths in non-sequential control structures. Sadiq and Orlowska [30] proposed a visual verification approach and algorithm with a set of graph reduction rules to discover structural conflicts in process models for given workflow modeling languages.
There are several research topics discussed in resource dimension, including resource allocation constraints [2, 3], resource availability [4], resource management [5] and resource modeling [6]. Senkul [2] developed an architecture to model and schedule workflow with resource allocation constraints and traditional temporal/causality constraints. Li [3] concluded that a correct workflow specification should have resource consistence. His algorithms can verify resource consistency and detect the potential resource conflicts for workflow specifications. Both Pinar and Hongchen extended workflow specifications with constraint descriptions. Liu [4]
proposed a three-level bottom-up workflow design method to effectively incorporate confirmation and compensation in case of failure. In Liu’s model, data resources are modeled as resource classes, and the only interface to a data resource is via a set of operations.
Current analysis techniques including above approaches pay little attention on the data-flow dimension, although the related analysis in data-flow dimension is very important since activities cannot be executed properly without sufficient data information. In the literature, there are two works in data-flow dimension found. Sadiq et al. [7] presented data flow validation issues in workflow modeling, including identifying requirements of data modeling and seven basic data validation problems: redundant data, lost data, missing data, mismatched data, inconsistent data, misdirected data, and insufficient data. However, there is no concrete verification procedure presented. Sun et al. [8－10] presented a data-flow analysis framework for detecting data-flow anomalies such as missing data, redundant data, and potential conflicts of data. In addition, they provided several analysis algorithms; however, the work is done only based on read and initial write data operations.