Linear Exact Repair in MDS Array Codes: A General Lower Bound and Its Attainability

Abstract: For an $(n,k,\ell)$ MDS array code over $\mathbb{F}q$, how small can the repair bandwidth and repair I/O be under linear exact repair? We study this question in the regime where the field size $q$, the redundancy $r=n-k$, and the sub-packetization level $\ell$ are fixed, while the code length $n$ varies, and we develop a geometric approach to this setting. Our starting point is an intrinsic reformulation of linear exact repair for MDS array codes in terms of subspace intersections and, for repair I/O, the projective point configurations induced by a parity-check realization. This viewpoint yields a simple projective counting argument establishing the general lower bound $$β{\mathrm{avg}},β{\max},γ{\mathrm{avg}},γ_{\max}\;\ge\;\ell(n-1)-\frac{q^{(r-1)\ell}-1}{q-1}$$ for linear exact repair of every $(n,k,\ell)$ MDS array code over $\mathbb{F}_q$ with redundancy $r=n-k\ge 2$. To our knowledge, this is the first lower bound of this form that applies to arbitrary redundancy $r\ge 2$ and sub-packetization level $\ell$. At first glance, the projective counting bound appears rather coarse and therefore unlikely to be attained. We prove that this intuition is correct whenever $r\ge 3$ and $\ell\ge 2$. For $r=2$, the picture changes completely. Using Desarguesian spreads from finite geometry, we construct MDS array codes that attain the bound over a broad interval of code lengths, up to the maximum possible length $q^{\ell}+1$, and do so simultaneously for both repair bandwidth and repair I/O. In the smallest nontrivial case $(r,\ell)=(2,2)$, we also prove a converse within the regular-spread model. Together, these results identify a uniform obstruction governing linear exact repair and show that, in the two-parity case, this obstruction is tight.

• The paper establishes a universal lower bound on repair bandwidth and I/O by linking subspace packing in projective spaces to finite-geometry limits.
• It proves that the bound is tight only for two-parity (r=2) codes while remaining strictly loose for higher redundancy regimes.
• The paper provides explicit optimal constructions using Desarguesian spreads, offering a practical blueprint for distributed storage systems.

Linear Exact Repair in MDS Array Codes: Lower Bounds and Attainability

Problem Setting and Motivation

The paper "Linear Exact Repair in MDS Array Codes: A General Lower Bound and Its Attainability" (2604.04519) investigates the intrinsic limitations and explicit constructions for the linear exact repair of Maximum Distance Separable (MDS) array codes, especially focusing on repair bandwidth and repair I/O when redundancy ($r$), sub-packetization level ($\ell$), and field size ($q$) are fixed. The central challenge is to determine how small the repair costs can be, expressed as the minimum total amount of symbols read and downloaded from helper nodes to repair a failed one, across all linear schemes and code realizations. This is especially relevant in distributed storage systems where efficient node-repair procedures are operationally critical.

Crucially, the authors frame the problem in the general regime where the code length $n$ varies, but $q$, $r$, and $\ell$ are held fixed. Prior works have focused either on asymptotics, specialized parameters, or particular code constructions (e.g., MSR codes or scalar Reed–Solomon codes), but no previously known lower bound in this general form captures the interplay between all four parameters for arbitrary redundancy and sub-packetization.

Intrinsic Subspace Reformulation of Linear Exact Repair

The analysis departs from traditional matrix-based repair descriptions by introducing an intrinsic geometric/combinatorial viewpoint. The column spaces of the parity-check blocks are framed as subspaces $\mathcal{H}_i \leq \mathbb{V}$, generating the code via a block-structured parity-check matrix. Linear exact repair of node $i$ is defined via the existence of a repair matrix $M$ with prescribed interactions with these subspaces, reformulated as the selection of a repair subspace $\ell$0 of $\ell$1 disjoint from $\ell$2.

Within this framework, the repair bandwidth for node $\ell$3 is associated with the aggregated dimensions $\ell$4 over the helper nodes, and repair I/O is governed by the number of projective points captured from the helper’s column sets. Thus, minimizing repair bandwidth/I/O reduces to packing as many helper-aligned subspace intersections inside a fixed-dimensional ambient subspace $\ell$5 as possible, subject to the $\ell$6-wise direct sum constraint among all node subspaces. This precise geometric language enables uniform counting arguments and connection to finite geometry.

The Projective Counting Lower Bound

The main theoretical result is a uniform lower bound for repair bandwidth and repair I/O for all $\ell$7 MDS array codes:

$\ell$8

This is obtained by considering the maximum number of pairwise-disjoint, nonzero $\ell$9-dimensional subspaces that can fit within the projective space associated to the repair subspace. The quantity $q$0 represents the maximal number of one-dimensional subspaces (projective points) inside an $q$1-dimensional space, reflecting the tightest that intersecting structures can be packed under the MDS constraint.

This lower bound is general: unlike prior work, it applies for any $q$2 and $q$3, not only for special cases or for large $q$4 asymptotics, and covers both average and worst-case scenarios for both repair metrics.

Tightness and Non-Attainability for Higher Redundancy

A significant technical contribution is the proof that this lower bound is not attainable except in the $q$5 (double-parity) case. For $q$6 and $q$7, the lower bound is always strictly loose for any code length permitted by the MDS property. This is established via a pair of combinatorial arguments: the first shows that equality would require more nodes than is feasible by the MDS direct-sum constraint, while the second establishes that no set of node subspaces can saturate the projective counting argument without exceeding this length bound.

Explicit Optimal Constructions for Two-Parity Codes

In contrast, for $q$8 (codes with two parity nodes), the lower bound becomes tight and is simultaneously attainable for both repair bandwidth and repair I/O. The core result demonstrates the existence, for all code lengths $q$9 with

$n$0

where $n$1, of $n$2 MDS array codes over $n$3 with

$n$4

The construction uses Desarguesian spreads—an extremal set system in finite geometry that precisely partitions the nonzero vectors of the underlying space—allowing for careful alignment of node subspaces and repair subspaces. The construction applies in a wide parameter range, up to the maximal possible code length $n$5. For the smallest nontrivial case $n$6, an exact converse is given under a regular-spread model, showing that the length threshold is not only sufficient but necessary for attainability in this regime.

Theoretical Implications and Connections

These results clarify that in the fixed-field, fixed-packetization regime, optimal linear repair is fundamentally constrained by finite-geometry parameters; the projective counting argument provides a universal obstruction that is tight only in the two-parity case. The identification of regular spreads (Desarguesian or projectively equivalent) as the only candidates for saturating the lower bound when $n$7 reinforces the deep connection between storage codes and combinatorial designs.

The strictness for $n$8 means that designers of distributed storage systems cannot hope, in general, to match the lower bound when using small redundancy and small sub-packetization. This impossibility is robust, independent of field size or code realization, and suggests that new approaches or relaxation of code properties (e.g., non-exact repair, non-linear repair, scaling $n$9 with $q$0) are necessary for further bandwidth/I/O reductions in regimes of practical interest.

Practical and Future Directions

Practically, the two-parity attainability is significant because many real-world systems (e.g., RAID-6 and its extensions) deploy double-parity codes for cost-efficient, reliable storage with manageable complexity. The explicit extremal constructions provided—in terms of projective design with Desarguesian spreads—guide practitioners in realizing codes that simultaneously minimize both repair bandwidth and I/O, subject to the MDS constraint, in this regime.

For higher redundancy, the negative result motivates the search for trade-offs beyond the strict linear MDS setting: for instance, exploring non-linear repair strategies, extending sub-packetization, or leveraging code locality in other ways. Additionally, the spread-based geometric approach may shed light on more general questions about code optimality, projective packings, and access-optimal code designs. Further research could investigate whether the lower bound is sharp for other code classes or in the presence of additional system constraints (e.g., degraded read-friendliness or helper-set restrictions).

Conclusion

This work rigorously settles the attainable limits of linear exact repair for MDS array codes with fixed redundancy, sub-packetization, and field size, providing a geometric lower bound valid in full generality. It is shown that this projective counting bound is never met except for codes with exactly two parities, where explicit constructions do achieve equality. Hence, the study delivers both a theoretical boundary for what is fundamentally possible and a practical blueprint for code construction in double-parity settings, while excluding the possibility of such optimality in higher redundancy regimes within the linear-MDS model (2604.04519).