Worst-case, information and all-blocks locality in distributed storage systems: An explicit comparison

Distributed storage systems often use erasure coding techniques to provide reliability while decreasing the storage overhead required by replication. Due to the drawbacks of standard MDS erasure-correcting codes, numerous coding schemes recently proposed for distributed storage systems target other metrics such as repair locality and repair andwidth. Unfortunately, these schemes are not always practical, and for most of them locality covers information data only. In this article, we compare three explicit linear codes for three types of locality: a Reed-Solomon code for worst-case locality, a recently proposed pyramid code for information locality and the Hamming code HAM, an optimal locally repairable code directly built from its generator matrix for all-blocks locality. We also provide an efficient way for repairing HAM and show that for the same level of storage overhead HAM provides faster encoding, faster repair and lower repair bandwidth than the other two solutions while requiring less than fifty lines of code.