Locality-Improved Repairable Fountain Codes For Distributed Storage Systems

We propose systematic fountain codes over GF(q) with the maximum likelihood (ML) decoding for distributed storage systems (DSS), which is called locality-improved repairable fountain codes (LIRFC). A generator matrix is constructed by entry-wise sampling and row-discarding, where the parameters are carefully determined such that locality is reduced while the ML decoding performance is not degraded. The degree distribution of the proposed codes is represented as a truncated and scaled binomial distribution with an optimized truncation threshold. Compared to the repairable fountain codes (RFC) that have logarithmic locality, the proposed codes achieve 5.97% to 22.44% less average locality than RFC, while having equivalent decoding failure probability and higher availability.