Partial replication with strong consistency

• Autor(a) principal: Matos, Tiago André Envia
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/147473
• topic: distributed systems; consistency; partial replication; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

In response to the increasing expectations of their clients, cloud services exploit geo-replication to provide fault-tolerance, availability and low latency when executing requests. However, cloud platforms tend to adopt weak consistency semantics, in which replicas may diverge in state independently. These systems offer good response times but at the disadvantage of allowing potential data inconsistencies that may affect user experience. Some systems propose to adopt solutions with strong consistency, which are not as efficient but simplify the development of correct applications by guaranteeing that all replicas in the system maintain the same database state. Therefore, it is interesting to explore a system that can offer strong consistency while minimizing its main disadvantage: the impact in performance that results from coordinating every replica in the system. A possible solution to reduce the cost of replica coordination is to support partial replication. Partially replicating a database allows for each server to only be responsible for a subset of the data - a partition - which means that when updating the database only some of replicas have to be synchronized, improving response times. In this dissertation, we propose an algorithm that implements a distributed replicated database that offers strong consistency with support for partial replication. To achieve strong consistency in a partially replicated scenario, our algorithm is in part based on the Clock-SI[10] research, which presents an algorithm that implements a multi-versioned database for strong consistency (snapshot-isolation) and performs the Two-Phase Commit protocol when coordinating replicas during updates. The algorithm is supported by an architecture that simplifies distributing partitions among datacenters and efficiently propagating operations across nodes in the same partition, thanks to the ChainPaxos[27] algorithm.