Section: New Results

Data driven systems

Web services

Participants : Blaise Genest, Loïc Hélouët.

This year, we considered transactional properties (ACID) for web services. In particular, we focused on the atomicity (A of ACID) property, obtained in case of a failure inside an atomic block through compensation of the executed actions of the block. To do so, logs need to be kept. We were interested in maintaining the maximal amount of privacy. We proposed modular algorithms [19] which maintain privacy between modules, with minimal information shared among modules, both in the logging and the compensation phases. Furthermore, each module logs a small number of information, such that the sum of all actions logged is guaranteed minimal. Last, modularity allows fast algorithms, as they need to consider only what happens in the module itself, and not the exact structure of its parent module nor of its sub-modules.

We also published results on our model of sessions systems [27] . This models allows for the modeling of distributed web-based systems that are running an arbitrary number of transactions among arbitrarily many participants. We have shown how simple restrictions can guarantee decidability of simple coverability properties, and then be used to detect violation of buisness rules such as conflict of interest, or a more complex property called the chinese wall.

We are currently considering new models that manage at the same time explicit workflows and structured data. This model can be seen as a combination of AXML  [46] and Petri nets.

An Artifact-centric Process Model

Participants : Éric Badouel, Loïc Hélouët, Christophe Morvan.

In [37] we present a purely declarative approach to artifact-centric case management systems, and a decentralization scheme for this model. Each case is presented as a tree-like structure; nodes bear information that combines data and computations. Each node belongs to a given stakeholder, and semantic rules govern the evolution of the tree structure, as well as how data values derive from information stemming from the context of the node. Stakeholders communicate through asynchronous message passing without shared memory, enabling convenient distribution.