Replicated convolutional codes: A design framework for repair-efficient distributed storage codes

Erasure-coded distributed storage systems can offer reliable storage services in a cost-effective manner. However, when disk failures occur in such systems, it is desirable to recreate the lost data with the help of surviving nodes to preserve the data redundancy. A key requirement during the recover process is to minimize the repair locality and computational complexity. We propose a simple framework for constructing storage codes that yield small repair locality and low repair complexity. The basic idea behind this framework is to take multiple instances of convolutional (tail-biting) codes, and then carefully arrange the coded symbols on storage nodes. The resultant codes enjoy the desirable repair-by-transfer property, and perform efficient repair by simple XOR operations. Moreover, we also evaluate the proposed codes atop an HDFS cluster testbed and compare the empirical performance with state-of-the-art repair-efficient storage codes.