Une architecture pour intégrer des composants de contrôle de la coopération dans un atelier distribué

Extended abstract

In this thesis we detail a new advanced transaction model that not only supports cooperation between activities through intermediate results exchanges while executing, but also allows them to be distributed and autonomous and to use possibly different cooperation schema to share their data. Our main objective was to decentralize the control of interactions towards the activities themselves, ie:

• each activity coordinates its own interactions with other activities,

• these controls performed locally implicitly ensure the synchronization of the whole system.

With that in mind, we defined two distributed corectness criteria: the D-serializability and the the DisCOO-serializability. They provide the same global properties than classical corectness criteria (serializability and COO-serializability) but they only rely on local controls performed by each transaction.

Besides the decentralization of the control of interactions, our second objective was to define mechanisms to let transactions negotiate the rules (cooperation schemas) they want to verify on their data exchanges.