Estudo comparativo entre replicação e erasure coding no armazenamento de dados em nuvem

Abstract

The amount and diversity of digitally generated data has increased exponentially, leading to the increasing use of cloud storage services due to their scalability, high performance, and convenience. However, storing large volumes of data poses significant challenges to the infrastructure, especially in terms of financial costs and performance. This study compares two data protection methods used in cloud storage systems: replication and erasure coding. Replication creates copies of the data in different hardware locations, aiming to increase availability and reduce bandwidth during transmission, but results in high storage costs. In contrast, erasure codes offer lower storage costs by fragmenting the data, and are widely adopted in large-scale systems such as Microsoft Azure Storage and Google Cloud Storage. However, they present a high computational cost due to encoding and decoding operations. This study investigates the performance of OpenStack Swift with different erasure codes, analyzing how replication and erasure coding policies affect read and write operations and space usage. The results indicate that 2-copy replication is more efficient for small files, while 3-copy replication excels for larger files. On the other hand, erasure codes with (4,2) configuration show better efficiency in terms of storage space, with performance comparable to or superior to replication under certain conditions. The detailed analysis of the experimental results produced provides insights for choosing the most appropriate storage policy, considering different contexts and needs.