An Efficient One-to-One Piggybacking Design for Distributed Storage Systems

Abstract

As a kind of erasure code, piggybacking has been applied in practice since it can significantly reduce the repair bandwidth of distributed storage systems. Currently, several efficient piggybacking designs have been proposed. In this paper, we propose a more efficient “one-to-one” piggybacking design (OOP) to further reduce the repair bandwidth. Different from the existing piggybacking designs, OOP adopts a simple encoding principle that one parity node only piggybacks symbols from one substripe. Particularly, OOP takes into account the efficient repair of systematic nodes and parity nodes simultaneously. It is shown that for OOP design, the optimal number of substripes is $(\sqrt {r-1}+r-1)$ , the average repair bandwidth ratio of systematic nodes can be as low as $\frac {2\sqrt {r-1}+1}{2\sqrt {r-1}+r}$ , and the average repair bandwidth ratio of parity nodes reaches $\left({\frac {\sqrt {r-1}+1}{r}+\frac {(r-1)^{2}-\sqrt {(r-1)^{3}}}{rk}}\right)$ . In contrast to the existing piggybacking designs, OOP can further reduce the repair bandwidth of both system nodes and parity nodes.