Use case induced relaxed transactional consistency models

The consistency models of in-memory data cluster and grid systems attempt to reconcile safe and comfortable programming on one side and efficient and highly concurrent program execution on the other side. Finding a way to maintain the consistency of replicated data in parallel and distributed programs that is convenient to use but still offers good performance is a complex challenge. The transactional memory approach promises to keep the programming effort low and still to provide safe and efficient execution. In many cases though transactional semantics are too rigid and create unnecessary transaction restarts. Our objective now is to partially relax the consistency model of the application while maintaining the strong transactional consistency as the base model. In this paper we present a platform that allows the coexistence of multiple consistency models, employing the transactional memory paradigm in a distributed fashion as the base consistency at the same time.