Replicação síncrona para sistemas de armazenamento chave-valor em memória primária

Abstract

In response to consistency and replication challenges found in Redis, this paper addresses the use of synchronous replication as a strategy to provide fully data consistency to a database, with the option of on disk data persistence and recovery support. In addition, this work sheds some light on the trade off between consistency and performance in replicated in-memory databases.

In this work, a proxy was developed using the Treplica synchronous replication framework, which implements the Paxos consensus algorithm and gives Redis the properties of data replication in a P2P model, data consistency, and supports the fail-recovery failure model. A benchmark was carried out comparing the proposed system with an application that uses Redis's native replication and failure recovery model, with the aim of understanding the trade-off between consistency and system performance. We have observed that the trade-off between eventual data consistency and fully data consistency in this application reduces performance by 8.4 to 12.4 times, as expected, but still delivers a high write rate, between 32.82 to 213.92 transactions per second in the tests performed.