Two-tone Shift-XOR Storage Codes

Storage codes using shift and XOR operations have been studied to achieve lower encoding and decoding computation costs, compared with the codes using large finite field operations. In this paper, we introduce a new class of shift-XOR codes using two-tone generator matrices, which generalize the existing increasing-difference generator matrices. Compared with the latter, our codes only have 1/3 to 1/2 storage overhead for practical cases, and have a decoding algorithm that preserves the desired properties. For two-tone shift-XOR codes, the reflected Vandermonde matrices achieve the smallest storage overhead; and for increasing-difference shift-XOR codes, the Vandermonde matrices achieve the smallest storage overhead. To verify the practical performance, we implement two-tone shift-XOR storage codes using C++ and compare the encoding/decoding throughput with the state-of-the-art implementation of Reed-Solomon codes. For certain practical cases, our codes can achieve from 50% to 100% higher encoding/decoding throughput than that of Reed-Solomon codes.