Cache transparente: uma solução para armazenamento de dados distribuídos voltada para cargas de trabalho intensivas em leituras

Abstract

In large-scale modern applications, users consume much more data than they create. This behavior results in data storage systems with workloads dominated by reads. Faced with this scenario, designers have adopted solutions such as primary copy replication and cache mechanisms. These solutions employ distributed data storage systems, which face major challenges to ensure data consistency. Currently, the literature lacks a formal and in-depth analysis of the consistency provided by these solutions and which problems arise from their use. This work presents a formal analysis that identifies the consistency problems presented by current solutions and introduces Cache-Through, a new solution capable of addressing such problems. The analysis of current solutions demonstrates that various consistency anomalies can happen. The design of Cache-Through is presented in detail and its consistency is formally demonstrated. The comparison between the solutions shows that Cache-Through is more consistent and more efficient in network usage than current solutions. The contributions made by this work allow designers to identify more precisely the consistency of their systems and provide them with a new solution to consider.