Description
Date depot: 1 janvier 1900
Titre: Just-Right Consistency
Directeur de thèse:
Marc SHAPIRO (LIP6)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Systèmes et réseaux
Resumé:
In a distributed system, data is distributed and replicated, in order to ensure availability and persistence despite high latencies and (unavoidable) failures.
This creates the issue of consistency: the famous 'CAP' impossibility result shows there is an inherent trade-off between fault tolerance, performance, and programmability. Strong consistency is very safe but is slow, expensive, and unavailable when network connections break; eventual consistency is fast and cheap but exposes programmers to bug-prone anomalies. Intermediate models exist but they are hard to understand. No single consistency model is appropriate for all uses.
We propose to develop a novel “Just-Right Consistency” (JRC) approach. As consistency aims to guarantee application invariants, we propose to tailor the distributed synchronisation mechanisms precisely to the needs of the specific application. It aims to provide the highest degree of availability that is compatible with the application requirements, and just enough consistency to ensure its invariants.
The JRC approach rests on advanced techniques from the distributed algorithms area and from application analysis and verification. It deconstructs the traditional consistency algorithms into fine-grain, efficiently composable orthogonal primitives. Based on an static and dynamic analysis of the application, the approach identifies the application's requirements in terms of synchronisation and data access. This analysis identifies any consistency anomalies or availability/performance bottlenecks, and accordingly proposes either to weaken a specific requirement or to strengthen a specific action.
Doctorant.e: Nair Sreeja