[Paper Review] New MDS codes with small sub-packetization and near-optimal repair bandwidth

Venkatesan Guruswami, Ankit Singh Rawat|arXiv (Cornell University)|Jul 31, 2016

Advanced Data Storage Technologies32 references3 citations

TL;DR

This paper presents a novel construction of MDS vector codes with small sub-packetization level ℓ = O(n−k) and near-optimal repair bandwidth—specifically, at most (1+1/t) times the cut-set bound for any integer t≥1. The codes achieve efficient exact repair via repair-by-transfer mechanisms, enabling practical deployment in distributed storage systems with minimal computational overhead at repair nodes.

ABSTRACT

An $(n, M)$ vector code $\mathcal{C} \subseteq \mathbb{F}^n$ is a collection of $M$ codewords where $n$ elements (from the field $\mathbb{F}$) in each of the codewords are referred to as code blocks. Assuming that $\mathbb{F} \cong \mathbb{B}^{\ell}$, the code blocks are treated as $\ell$-length vectors over the base field $\mathbb{B}$. Equivalently, the code is said to have the sub-packetization level $\ell$. 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 $M = |\mathbb{B}|^{k\ell}$, the repair bandwidth of an MDS vector code is lower bounded by $\big(\frac{n - 1}{n - k}\big)\cdot \ell$ symbols (over the base field $\mathbb{B}$) 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 $\ell = (n-k)^{\lceil{{n}/{(n-k)}} ceil}$ are known in the literature [23, 35]. 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 $\ell = O(n - k)$ 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 $\mathbb{B}$.

Motivation & Objective

• To address the trade-off between sub-packetization level ℓ and repair bandwidth in MDS codes for distributed storage systems.
• To construct MDS codes that achieve near-optimal repair bandwidth while keeping ℓ as small as possible, ideally O(n−k).
• To enable practical deployment by supporting repair-by-transfer, minimizing computation at contacted nodes during repair.
• To generalize the construction so that repair bandwidth approaches the cut-set bound at the cost of slightly increased sub-packetization.

Proposed method

• The construction uses a combinatorial design based on structured parity-check matrices with specific sparsity patterns to enable efficient repair.
• It partitions the code symbols into stages and uses linear constraints to recover missing symbols by downloading only a limited number of symbols from other nodes.
• The repair process is divided into two stages: stage 1 downloads (n−1)r^{t−1} symbols, and stage 2 downloads at most (r−1)r^{t−1}⌊s/t⌋ symbols, where r = n−k.
• The repair bandwidth is bounded by (1+1/t)(n−1)/r · ℓ, which is (1+1/t) times the cut-set bound.
• The code is linear over the base field 𝔹 and relies on the combinatorial structure of the parity-check matrix to ensure the MDS property.
• The construction generalizes to allow repair bandwidth approaching the cut-set bound by increasing the sub-packetization level to (d−k+1)^t for general d.

Experimental results

Research questions

• RQ1 Can MDS codes be constructed with sub-packetization level ℓ = O(n−k) while maintaining near-optimal repair bandwidth?
• RQ2 How can repair-by-transfer be achieved in MDS codes with small ℓ to reduce computational cost during node repair?
• RQ3 What is the trade-off between sub-packetization level and repair bandwidth when approaching the cut-set bound?
• RQ4 Can the construction be generalized to support repair with d < n−1 contacted nodes, not just d = n−1?
• RQ5 Is it possible to reduce the required base field size while preserving the MDS property and repair efficiency?

Key findings

• The proposed codes achieve a sub-packetization level ℓ = O(n−k), which is the smallest possible for MDS codes with near-optimal repair bandwidth.
• The repair bandwidth is at most (1+1/t) times the cut-set bound, with t being a design parameter that controls the trade-off.
• For t=1, the repair bandwidth is at most 2 times the cut-set bound, which is a significant improvement over prior constructions with small ℓ.
• The construction supports repair-by-transfer, meaning no computation is needed at the contacted nodes during repair, enhancing efficiency.
• The codes are linear over the base field 𝔹 and maintain the MDS property due to the carefully designed combinatorial structure of the parity-check matrix.
• The method generalizes to arbitrary d < n−1, yielding codes with sub-packetization level (d−k+1)^t and repair bandwidth approaching the cut-set bound as t increases.