Responsive Workflows: Design, Execution and Analysis of Interruption Policy Models

Business processes form the backbone of all business operations, and workflow technology has enabled companies to gain significant productivity benefits through the automatic enactment of routine, repetitive processes. Process automation can be achieved by encoding the business rules and procedures into the applications, but capturing the process logic in a graphical workflow model allows the process to be specified, validated and ultimately maintained by business analysts with limited technical knowledge. The process models can also be automatically verified at design-time to detect structural issues such as deadlock and ensure correct data flow during process execution. These benefits have resulted in the success of workflow technology in a variety of industries, although workflows are often criticised for being too rigid, particularly in light of their recent deployment in collaborative applications such as e-business. Generally, many events can impact on the execution of a workflow process. Initially, the workflow is triggered by an external event (for example, receipt of an order). Participants then interact with the workflow system through the worklist as they perform constituent tasks of the workflow, driving the progression of each process instance through the model until its completion. For traditional workflow processes, this functionality was sufficient. However, new generation 'responsive' workflow technology must facilitate interaction with the external environment during workflow execution. For example, during the execution of an 'order to cash' process, the customer may attempt to cancel the order or update the shipping address. We call these events 'interruptions'. The potential occurrence of interruptions can be anticipated but, unlike the other workflow events, they are never required to occur in order to successfully execute any process instance. Interruptions can also occur at any stage during process execution, and may therefore be considered as 'expected, asynchronous exceptions' during the execution of workflow processes. Every interruption must be handled, and the desired reaction often depends on the situation. For example, an address update may not be permitted after a certain point, where this point depends on the customer type, and a shipping charge or refund may be applicable, depending on the original and new delivery region. Therefore, a set of rules is associated with each interruption, such that if a condition is satisfied when the event occurs, a particular action is to be performed. This set of rules forms a policy to handle each interruption. Several workflow systems do facilitate the automatic enforcement of 'exception handling' rules and support the reuse of code fragments to enable the limited specification and maintenance of rules by non-technical users. However, this functionality is not represented in a formal, intuitive model. Moreover, we argue that inadequate consideration is given to the verification of the rules, with insufficient support provided for the detection of issues at design-time that could hinder effective maintenance of the process logic or interfere with the interruption handling functionality at run-time. This thesis presents a framework to capture, analyse and enforce interruption process logic for highly responsive processes without compromising the benefits of workflow technology. We address these issues in two stages. In the first stage, we consider that the reaction to an interruption event is dependent on three factors: the progress of the process instance with respect to the workflow model, the values of the associated case data variables at the time at which the event occurs, and the data embedded in the event. In the second stage, we consider that the reaction to each interruption event may also depend on the other events that have also been detected, that is, we allow interruptions to be defined through event patterns or complex events. We thus consider the issues of definition, analysis and enactment for both 'basic' and 'extended' interruption policy models. First, we introduce a method to model interruption policies in an intuitive but executable manner such that they may be maintained without technical support. We then address the issue of execution, detailing the required system functionality and proposing a reference architecture for the automatic enforcement of the policies. Finally, we introduce a set of formal, generic correctness criteria and a verification procedure for the models. For extended policy models, we introduce and compare two alternative execution models for the evaluation of logical expressions that represent interruption patterns. Finally, we present a thorough analysis of related verification issues, considering both the system and user perspectives, in order to ensure correct process execution and also provide support for the user in semantic validation of the interruption policies