Relaxed data consistency with CONStanza

TLDR

Data replication enhances fault tolerance and availability, yet most grid replication tools assume read‑only data, whereas many applications require updates to replicated items. The authors introduce a replica consistency service that permits updates in a single‑master setting using lazy synchronization. The service supports updates to heterogeneous relational databases and flat files, keeping remote replicas partially synchronized through lazy consistency. The authors demonstrate the service’s effectiveness in a typical application use case, highlighting its practical usefulness.

Abstract

Data replication is an important aspect in a data grid for increasing fault tolerance and availability. Many grid replication tools or middleware systems deal with read-only files which imply that replicated data items are always consistent. However, there are several applications that do require updates to existing data and the respective replicas. In this article, we present a replica consistency service that allows for replica updates in a single-master scenario with lazy update synchronisation. The system allows for updates of (heterogeneous) relational databases, and it is designed to support flat files as well. It keeps remote replicas synchronised and partially ("lazily") consistent. We report on the design and implementation of a novel "relaxed" replica consistency service and show its usefulness in a typical application use case.