The workflow models have been essentially operation-centric for many years, ignoring almost completely the data aspects. Recently, a new paradigm of data-centric workflows, called business artifacts , has been introduced by Nigam and Caswell. We follow this approach and propose a model where artifacts are XML documents that evolve in time due to interactions with their environment, i.e. human users or Web services. This paper proposes the AXART system as a distributed platform for collaborative work that harnesses the power of our model. We will illustrate AXART with an example taken from the movie industry. Indeed, applying for a role in a film is a typical collaborative process that involves various participants, inside and outside the film company. The demonstration scenario considers both standard workflow process and dynamic workflow modifications, based on two extension mechanisms: workflow specialization and workflow exception. The workflows, modeled using artifacts, are supported by the AXART system by combining techniques specific to active documents, like view maintenance, with security techniques to manage access rights.

Many Web applications are based on dynamic interactions between Web components exchanging flows of information. Such a situation arises for instance in mashup systems or when monitoring distributed autonomous systems. Our work is in this challenging context that has generated recently a lot of attention; see Web 2.0. We introduce the axlog formal model for capturing such interactions and show how this model can be supported efficiently. The central component is the axlog widget defined by one tree-pattern query or more, over an active document (in the Active XML style) that includes some input streams of updates. A widget generates a stream of updates for each query, the updates that are needed to maintain the view corresponding to the query. We exploit an array of known technologies: datalog optimization techniques such as Differential or MagicSet, constraint query languages, and efficient XML filtering (YFilter). The novel optimization technique we propose is based on fundamental new notions: a relevance (different than that of MagicSet), satisfiability and provenance for active documents. We briefly discuss an implementation of an axlog engine, an application that we used to test the approach, and results of experiments.

ManyWeb applications are based on dynamic interactions between Web components exchanging flows of information. Such a situation arises for instance in mashup systems [22] or when monitoring distributed autonomous systems [6]. This is a challenging problem that has generated recently a lot of attention; see Web 2.0 [38]. For capturing interactions between Web components, we use active documents interacting with the rest of the world via streams of updates. Their input streams specify updates to the document (in the spirit of RSS feeds), whereas their output streams are defined by queries on the document. In most of the paper, the focus is on input streams where the updates are only insertions, although we do consider also deletions.

Towards a data-centric workflow approach, we introduce an artifact model to capture data and workflow management activities in distributed settings. The model is built on Active XML, i.e., XML trees including Web service calls. We argue that the model captures the essential features of business artifacts as described informally in [A. Nigam and N. Caswell (2003)] or discussed in [R. Hull (2008)]. To illustrate, we briefly consider the monitoring of distributed systems and the verification of temporal properties for them.

Observing highly dynamic peer-to-peer systems is essential for many applications such as fault management or business processing. We demonstrate P2PMonitor, a P2P system for monitoring such systems. Alerters deployed on the monitored peers are designed to detect particular kinds of local events. They generate streams of XML data that form the primary sources of information for P2PMonitor. The core of the system is composed of processing components implementing the operators of an algebra over data streams. From a user viewpoint, monitoring a P2P system can be as simple as querying an XML document. The document is an ActiveXML document that aggregates a (possibly very large) number of streams generated by alerters on the monitored peers. Behind the scene, P2PMonitor compiles the monitoring query into a distributed monitoring plan, deploys alerters and stream algebra processors and issues notifications that are sent to users. The system functionalities are demonstrated by simulating the supply chain of a large company.