[Paper Review] MDS code constructions with small sub-packetization and near-optimal repair bandwidth

Venkatesan Guruswami, Ankit Singh Rawat|arXiv (Cornell University)|Jan 16, 2017

Advanced Data Storage Technologies14 citations

TL;DR

This paper presents a novel construction of MDS vector codes with minimal sub-packetization level e = O(n − k) and repair bandwidth within a factor of two of the cut-set bound. The method leverages linear algebraic structures over a base field to achieve near-optimal repair efficiency while maintaining small sub-packetization, offering a significant improvement over prior constructions in terms of practical storage system trade-offs.

ABSTRACT

An (n, M) vector code C ⊆ 𝔽n is a collection of M codewords where n elements (from the field 𝔽) in each of the codewords are referred to as code blocks. Assuming that 𝔽 ≅ 𝔹e, the code blocks are treated as e-length vectors over the base field 𝔹. Equivalently, the code is said to have the sub-packetization level e. This paper addresses the problem of constructing MDS vector codes which enable exact reconstruction of each code block by downloading small amount of information from the remaining code blocks. The repair bandwidth of a code measures the information flow from the remaining code blocks during the reconstruction of a single code block. This problem naturally arises in the context of distributed storage systems as the node repair problem [4]. Assuming that M = |𝔹|ke, the repair bandwidth of an MDS vector code is lower bounded by ((n − 1)/(n − k))· e symbols (over the base field 𝔹) which is also referred to as the cut-set bound [4]. For all values of n and k, the MDS vector codes that attain the cut-set bound with the sub-packetization level e = (n − k)⌈n/(n − k)⌉ are known in the literature [23,36].This paper presents a construction for MDS vector codes which simultaneously ensures both small repair bandwidth and small sub-packetization level. The obtained codes have the smallest possible sub-packetization level e = O(n − k) for an MDS vector code and the repair bandwidth which is at most twice the cut-set bound. The paper then generalizes this code construction so that the repair bandwidth of the obtained codes approach the cut-set bound at the cost of increased sub-packetization level. The constructions presented in this paper give MDS vector codes which are linear over the base field 𝔹.

Motivation & Objective

• To address the challenge of minimizing sub-packetization level in MDS vector codes while maintaining low repair bandwidth for efficient node repair in distributed storage.
• To construct MDS codes that simultaneously achieve small sub-packetization e = O(n − k) and repair bandwidth within a factor of two of the theoretical cut-set bound.
• To generalize the construction so that repair bandwidth approaches the cut-set bound at the cost of higher sub-packetization, enabling tunable trade-offs.
• To ensure the codes are linear over the base field 𝔽, facilitating practical implementation in storage systems.

Proposed method

• The construction uses linear algebraic techniques over a base field 𝔽 to define codewords as vectors of length n, with each block being an e-length vector over 𝔽.
• Sub-packetization level e is set to O(n − k), minimizing the number of sub-packets per block to reduce complexity.
• The repair process is designed to reconstruct any single code block by downloading a minimal amount of data from the remaining n−1 blocks, achieving bandwidth within a factor of two of the cut-set bound.
• A generalized version of the code is introduced where sub-packetization increases to allow repair bandwidth to approach the cut-set bound asymptotically.
• The codes are explicitly constructed to be linear over the base field 𝔽, ensuring compatibility with standard linear coding techniques.

Experimental results

Research questions

• RQ1 Can MDS vector codes be constructed with sub-packetization level e = O(n − k) while maintaining low repair bandwidth?
• RQ2 What is the minimal achievable repair bandwidth for such codes with small sub-packetization?
• RQ3 Can the repair bandwidth be made arbitrarily close to the cut-set bound by allowing a controlled increase in sub-packetization?
• RQ4 How can linear MDS codes be systematically designed to balance sub-packetization and repair efficiency?

Key findings

• The constructed MDS codes achieve the smallest possible sub-packetization level e = O(n − k) for any MDS vector code with near-optimal repair bandwidth.
• The repair bandwidth of the proposed codes is at most twice the cut-set bound, significantly improving upon prior constructions with comparable sub-packetization.
• By increasing sub-packetization, the repair bandwidth of the generalized codes approaches the cut-set bound, enabling a tunable trade-off between complexity and efficiency.
• All codes are linear over the base field 𝔽, ensuring compatibility with practical distributed storage implementations.