Dynamic Restructuring of Transactional Work ow Activities: A Practical Implementation Method

We describe a practical method to implement the dynamic restructuring of distributed transactional activities with several main advantages. First, it guarantees the correctness of dynamic restructuring operations as well as that of ongoing transac-tional activities. Second, it is general enough to accommodate a wide range of workkow activities' dynamic restructuring requirements. Third, it facilitates the deployment of dynamic restructuring operations on production systems. Our method combines adaptors, microprotocols, and incremental specialization, building modularly on the system components (e.g., RTF 1], OCP 28], MARS 4]) that we previously developed to support distributed extended transaction processing 19]. A prototype implementation based on this method is built on top of Transarc Encina, a commercial online transaction processing (OLTP) monitor. To our knowledge, this is the rst implementation of dynamic restructuring on top of commercial OLTP systems. 1 Introduction Many of today's business workkows execute in distributed environments, exhibit dynamic structure, and involve sophisticated interaction dependencies, making them vulnerable to uncertain duration and unpredictable development. Handling dynamic transactional workkows in practice is considered one of the challenging open problems in workkow research 24]. In order to optimize the structure of business workkows and their execution performance in anticipation of various uncertainties, we have introduced a family of operations (e.g., Activity-Split and Activity-Join) for dynamic restructuring of ongoing workkow activities 16]. These activity restructuring operations build on a formal notion of validity so that they guarantee a rich set of correctness criteria (that extend serializability) of the resulting distributed, dynamic, and sophisticated new activities. Despite the conceptual simplicity of these activity restructuring operations, their implementation on practical systems is challenging for a variety of reasons: (1) these restructuring operations need to be carried out dynamically and eeciently while activity instances are executing; (2) activity instances have trans-actional properties to be met in addition to their normal