Improving the scalability of geo-replication with reservations

Mahsa Najafzadeh (1) , Marc Shapiro (1) , Valter Balegas (2) , Nuno Preguiça (2)

1 Regal - Large-Scale Distributed Systems and Applications
2 CITI - Centro de Investigação em Informática e Tecnologias da Informação

Geo-replicated systems improve performance and fault tolerance by replicating data on sites in different physical locations. These systems often eschew guaranteeing strong consistency because of performance loss and scalability and instead choose eventually consistency. Although eventual consistency improves performance especially in large scale but it might violate system invariants. In this work, we exploit reservation techniques to strengthen eventual consistency, by adding safety guarantees. We define a consistency model called RPB that takes the advantages of eventual consistency while providing stronger guarantees, including causality and safety properties.

Mots clés: commutativity in distributed systems; conflict-free replicated data type (CRDT); synchronisation; replication and consistency

Domaines: Autre [cs.OH]; Calcul parallèle, distribué et partagé [cs.DC]

Mahsa Najafzadeh, Marc Shapiro, Valter Balegas, Nuno Preguiça. Improving the scalability of geo-replication with reservations. DCC 2013 : Workshop on Distributed Cloud Computing, Dec 2013, Dresden, Germany. pp.441-445, ⟨10.1109/UCC.2013.87⟩. ⟨hal-00932657⟩