Analysis of Recovery Characteristics of Erasure Coding and Replication

Abstract

The replication has been widely used to ensure data availability in a distributed file system. In addition, the erasure coding has been adopted to overcome a problem of space efficiency in the replication. However, the erasure coding has a number of performance degradation factors. In particular, the replication requires only replacement of replica and original data, whereas the erasure coding requires reading distributed data in many nodes and restoration of original data through decoding. In this paper, a number of characteristics related to fault recovery are studied prior to identifying an efficient restoration in the erasure coding. Through this, a number of characteristics that are considered when introducing the erasure coding in the distributed file system are derived. In particular, problems and considerations of the erasure coding in relation to fault recovery are verified.