Consensus-Based Management of Distributed and Replicated Data

Atomic Broadcast and Atomic Commitment are fundamental problems that have to be solved when managing distributed/replicated data. This short note aims at showing that solutions to these problems can benefit from results associated with the Consensus problem. Such an approach helps gain a better insight into distributed/replicated data management problems. More precisely, this note addresses one of the most important issues one is faced to when designing distributed/replicated data management protocols, namely, their Non-Blocking property. This property stipulates that the crash of nodes participating in a protocol must not prevent the non-crashed nodes from terminating the protocol execution. Results from the Consensus study allow to know the minimal assumptions a system must satisfy in order its distributed/replicated data management protocols be non-blocking despite process crash and asynchrony.