MDS codes with low repair complexity for distributed storage networks

In this paper, we propose two new constructions of maximum distance separable (MDS) codes with low repair complexity for distributed storage networks. For both constructions, the encoded symbols are obtained by first treating the message vector as a linearized polynomial and then evaluating it at carefully chosen points. Compared to traditional MDS codes, data repair for our codes does not have to decode the entire original message, but only forms linear combinations of available encoded symbols. This linear repair complexity is attractive for applications with a large amount of data repair.