Code Constructions for Distributed Storage With Low Repair Bandwidth and Low Repair Complexity

TL;DR

This paper introduces a new family of erasure codes for distributed storage that significantly reduce repair bandwidth and complexity by combining fault-tolerant MDS-based codes with piggyback modifications, optimizing repair procedures.

Contribution

The paper proposes a novel two-class code construction that balances fault tolerance with low repair bandwidth and complexity, improving upon existing codes like MDS, Zigzag, and Pyramid codes.

Findings

Achieves better repair bandwidth than MDS and related codes.

Reduces repair complexity compared to traditional codes.

Demonstrates numerical improvements in repair efficiency.

Abstract

We present the construction of a family of erasure correcting codes for distributed storage that achieve low repair bandwidth and complexity at the expense of a lower fault tolerance. The construction is based on two classes of codes, where the primary goal of the first class of codes is to provide fault tolerance, while the second class aims at reducing the repair bandwidth and repair complexity. The repair procedure is a two- step procedure where parts of the failed node are repaired in the first step using the first code. The downloaded symbols during the first step are cached in the memory and used to repair the remaining erased data symbols at minimal additional read cost during the second step. The first class of codes is based on MDS codes modified using piggybacks, while the second class is designed to reduce the number of additional symbols that need to be downloaded to repair the remaining erased symbols. We numerically show that the proposed codes achieve better repair bandwidth compared to MDS codes, codes constructed using piggybacks, and local reconstruction/Pyramid codes, while a better repair complexity is achieved when compared to MDS, Zigzag, Pyramid codes, and codes constructed using piggybacks.