Author(s): Cabrita, Gonçalo Mendes
Date: 2017
Persistent ID: http://hdl.handle.net/10362/21766
Origin: Repositório Institucional da UNL
Subject(s): Non-uniform Replication; Partial Replication; Replicated Data Types; Eventual Consistency; Key-Value Stores; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática
A large number of web applications/services are supported by applications running in cloud computing infrastructures. Many of these application store their data in georeplicated key-value stores, that maintain replicas of the data in several data centers located across the globe. Data management in these settings is challenging, with solutions needing to balance availability and consistency. Solutions that provide high-availability, by allowing operations to execute locally in a single data center, have to cope with a weaker consistency model. In such cases, replicas may be updated concurrently and a mechanism to reconcile divergent replicas is needed. Using the semantics of data types (and operations) helps in providing a solution that meets the requirements of applications, as shown by conflict-free replicated data types. As information grows it becomes difficult or even impossible to store all information at every replica. A common approach to deal with this problem is to rely on partial replication, where each replica maintains only part of the total system information. As a consequence, each partial replica can only reply to a subset of the possible queries. In this thesis, we introduce the concept of non-uniform replication where each replica stores only part of the information, but where all replicas store enough information to answer every query. We apply this concept to eventual consistency and conflict-free replicated data types and propose a set of useful data type designs where replicas synchronize by exchanging operations. Furthermore, we implement support for non-uniform replication in AntidoteDB, a geo-distributed key-value store, and evaluate the space efficiency, bandwidth overhead, and scalability of the solution.
