Maximally Recoverable Codes: the Bounded Case

Modern distributed storage systems employ Maximally Recoverable codes that aim to balance failure recovery capabilities with encoding/decoding efficiency tradeoffs. Recent works of Gopalan et al [SODA 2017] and Kane et al [FOCS 2017] show that the alphabet size of grid-like topologies of practical interest must be large, a feature that hampers decoding efficiency. To bypass such shortcomings, in this work we initiate the study of a weaker version of recoverability, where instead of being able to correct all correctable erasure patterns (as is the case for maximal recoverability), we only require to correct all erasure patterns of bounded size. The study of this notion reduces to a variant of a combinatorial problem studied in the literature, which is interesting in its own right. We study the alphabet size of codes withstanding all erasure patterns of small (constant) size. We believe the questions we propose are relevant to both real storage systems and combinatorial analysis, and merit further study.