Flexible modeling and execution of choreographies

Abstract

Approaches to address domain specific problems often share overlapping requirements but typically satisfy them in a unique manner for example using service-oriented concepts. The notion of Collaborative, Dynamic & Complex (CDC) systems has been proposed in literature to address the requirements of application domains such as eScience and Collective Adaptive Systems in a unified, generic manner. CDC systems are characterized by dealing with potentially large amounts of data and/or participating applications which engage in complex interactions specified by some collaboration protocol. Furthermore, the need for adaptation mechanisms is a common requirement and users from these application domains are typically no IT experts. The choreography concept originally known from collaborations in the business domain captures the interaction between independent parties from a global perspective. Each party is denoted as a choreography participant, which is implemented by a workflow or a service. This concept provides a way to model and execute for example complex eScience experiments involving multiple scientific fields, scientific methods, and time and/or length scales as a set of coupled workflows. However, typical choreography concepts as described in literature do not provide the desired level of flexibility and ease of use in both modeling and execution to address the requirements of users in CDC system application domains such as eScience. Thus, existing choreography concepts have to be considerably extended by introducing the Model-as-you-go for Choreographies approach in this thesis as a central notion providing capabilities for the flexible modeling and execution of choreographies. In the context of this approach, we provide a concept for fostering reuse in choreography modeling in the form of so-called choreography fragments. Such fragments can be extracted from existing and inserted into new choreography models in order to save time as well as reuse established and approved logic by inexperienced modelers in a less error-prone manner. Furthermore, we provide support for the user-driven control of the complete choreography life cycle. This effectively allows users to automatically deploy the workflow models implementing a choreography as well as starting, pausing, resuming, and terminating a choreography instance, which is formed through the collective execution of workflow instances. Most importantly, the underlying complexity of managing a set of coupled workflow instances is completely hidden from the users. Additional flexibility is given by a concept that allows to re-run already executed choreography logic in order to enforce the convergence of a calculation towards a particular result or to react to errors with parameter changes. The proposed concepts are implemented in a message-based system, the ChorSystem, which is able to handle the challenges of choreography life cycle management from deployment, to run time control and the re-run of logic. Furthermore, the modeling and run time monitoring are integrated into one graphical tool supporting the seamless transition from modeling to execution of choreographies. The concepts, their supporting algorithms, and the prototypical ChorSystem are validated by a set of case studies from different CDC system application domains and evaluated by performance measurements showing the practical applicability.