HFR code: a flexible replication scheme for cloud storage systems

Fractional repetition (FR) codes are a family of repair-efficient storage codes that provide exact and uncoded node repair at the minimum bandwidth regenerating point. The advantageous repair properties are achieved by a tailor-made two-layer encoding scheme which concatenates an outer maximum-distanceseparable (MDS) code and an inner repetition code. In this paper, we generalize the application of FR codes and propose heterogeneous fractional repetition (HFR) code, which is adaptable to the scenario where the repetition degrees of coded packets are different. We provide explicit code constructions by utilizing group divisible designs, which allow the design of HFR codes over a large range of parameters. The constructed codes achieve the system storage capacity under random access repair and have multiple repair alternatives for node failures. Further, we take advantage of the systematic feature of MDS codes and present a novel design framework of HFR codes, in which storage nodes can be wisely partitioned into clusters such that data reconstruction time can be reduced when contacting nodes in the same cluster. Index Terms Cloud storage systems, regenerating codes, repair bandwidth, uncoded repair, combinatorial designs. A preliminary version of this work was presented at the 2014 National Conference on Communications [1]. B. Zhu is with the School of Electronic and Computer Engineering, Peking University, Shenzhen 518055, P. R. China (e-mail: [email protected]). H. Li is with the Institute of Big Data Technologies and Shenzhen Engineering Lab of Converged Networks Technology, Peking University Shenzhen Graduate School, Shenzhen 518055, P. R. China (e-mail: [email protected]). K. W. Shum is with the Institute of Network Coding, The Chinese University of Hong Kong, Shatin, Hong Kong (e-mail: [email protected]). S.-Y. R. Li is with the University of Electronic Science and Technology of China, Chengdu 611731, P. R. China and Institute of Network Coding, The Chinese University of Hong Kong, Shatin, Hong Kong (e-mail: [email protected]). Corresponding author. September 15, 2015 DRAFT