Towards optimisation of replicated erasure codes for efficient cooperative repair in cloud storage systems

Abstract

The study of erasure codes in distributed storage systems has two aspects: one is to reduce the data redundancy and the other one is to save the bandwidth cost during repair process. Repair-efficient codes are investigated to improve the repair performance. However the researches are mostly in theoretical stage and hardly applied in the practical distributed storage systems like cloud storage. In this paper, we present a unified framework to describe some repair-efficient regenerating codes in order to reduce the bandwidth cost in regenerating the lost data. We build an evaluation system to measure the performance of these codes during file encoding, file decoding and individual failure repairing with given feasible parameters. By the experimental comparison and analysis, we validate that the repair-efficient regenerating codes can significantly save much more repair time than traditional erasure codes during repair process at the same storage cost; in particular, some replication-based erasure codes can perform better than others in some certain cases. Our experiments can help researchers to decide which kind of erasure codes to use in building distributed storage systems.