A Practical Study of Exact - BASIC Codes at the MSR Point in Distributed Storage Systems

Regenerating codes have been proven a class of optimal distributed storage codes in the tradeoff between storage capacity and repair bandwidth. However, existing regenerating codes rely on expensive computations such as finite field multiplication. The high coding complexity makes regenerating codes unsuitable for practical distributed storage systems. BASIC codes, standing for Binary Addition and Shift Implementable Convolutional codes, are proposed to reduce the computational complexity, as well as to keep the benefits of regenerating codes. In this paper, we implement an exact-repair BASIC code at the minimum-storage (MSR) point in a practical distributed storage system and compare it to Cauchy Reed-Solomon (CRS) code atop a cluster testbed with 20 storage nodes. The results show that minimum-storage BASIC code outperforms CRS code in terms of computational complexity and achieves a significant reduction both of repair bandwidth and disk I/O.