On the Impact of the Data Redundancy Strategy on the Recoverability of Friend-to-Friend Backup Systems

Marcelo Iury S . OliveiraWalfredo CirneFrancisco V. BrasileiroDalton Serey Guerrero

Social network-based systems, also known as Friend-to-Friend (F2F) systems, are a promising approach to develop backup solutions that provide high reliability with a much lower consumption of bandwidth and storage than P2P ones. F2F backup systems can use two data redundancy strategies to handle peer failure events, namely: replication and erasure coding. In this paper we evaluate the use of these alternatives to handle failures in F2F backup systems. The assessment is conducted using a new metric named recoverability slowdown. The proposed metric represents how efficiently one can restore the data lost due to the occurrence of failures. Our aim is to determine which data redundancy scheme constitutes a more balanced solution in terms of recoverability slowdown under different network bandwidth capacities and storage overhead levels. The simulations we have performed indicate that an increase on the storage overhead leads to better values for the backup recoverability slowdown when using the replication technique, while it leads to worse values when using the erasure coding technique. Moreover, using enough redundancy, replication can achieve results that are close to the optimal case for recoverability slowdown, while the best performance achieved by erasure coding is not larger than 87% of the optimal case. Nevertheless, for relatively low values of storage overhead, erasure coding outperforms replication.

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: