[Paper Review] Construction of Partial MDS (PMDS) and Sector-Disk (SD) Codes with Two Global Parity Symbols

Mario Blaum, James S. Plank+2 more|arXiv (Cornell University)|Jan 19, 2014

Advanced Data Storage Technologies4 citations

TL;DR

This paper presents a general algebraic construction for Partial-MDS (PMDS) and Sector-Disk (SD) codes with two global parity symbols, using finite fields and primitive elements. The method ensures maximal recoverability for any number of failed disks and up to two additional erased sectors, requiring only a modest field size, thus solving an open problem in erasure code design for modern storage systems.

ABSTRACT

Partial MDS (PMDS) codes are erasure codes combining local (row) correction with global additional correction of entries, while Sector-Disk (SD) codes are erasure codes that address the mixed failure mode of current RAID systems. It has been an open problem to construct general codes that have the PMDS and the SD properties, and previous work has relied on Monte-Carlo searches. In this paper, we present a general construction that addresses the case of any number of failed disks and in addition, two erased sectors. The construction requires a modest field size. This result generalizes previous constructions extending RAID~5 and RAID~6.

Motivation & Objective

• To address the open problem of constructing general PMDS and SD codes with two global parity symbols for arbitrary numbers of failed disks and two additional erased sectors.
• To eliminate reliance on Monte-Carlo searches by providing a deterministic, algebraic construction.
• To generalize prior constructions for RAID 5 and RAID 6 by extending them to support broader failure patterns.
• To minimize field size requirements while ensuring correctness under all valid erasure patterns.
• To provide a unified framework applicable to both PMDS and SD codes with two extra parity symbols.

Proposed method

• Uses finite fields $GF(2^w)$ with a primitive element $\alpha$ whose order exceeds the total number of array elements to ensure invertibility of key expressions.
• Constructs a parity-check matrix $H$ for SD codes with $m$ parity columns and two additional global parity symbols, using structured row blocks based on powers of $\alpha$.
• Implements a modified parity-check matrix $H'$ for PMDS codes by introducing a shift parameter $N = (m+1)(n-m-1)+1$ to ensure full rank under all erasure patterns.
• Employs minimal polynomials $f_\alpha(x)$ and polynomial rings $M_p(x)$ to generalize the construction beyond prime fields.
• Ensures invertibility of critical field expressions by bounding the exponent differences within the field order, preventing zero results.
• Validates correctness via algebraic proofs showing that all required erasure patterns are decodable under the constructed code structure.

Experimental results

Research questions

• RQ1 Can a deterministic, algebraic construction be developed for $(m;2)$ PMDS and SD codes that avoids Monte-Carlo search methods?
• RQ2 What field size is sufficient to guarantee correct decoding for any combination of $m$ failed disks and two additional erased sectors?
• RQ3 How can the construction be generalized to support arbitrary numbers of rows and columns while maintaining maximal recoverability?
• RQ4 What algebraic conditions ensure that the parity-check matrix remains full rank under all valid erasure patterns?
• RQ5 Can the same framework be adapted to support both PMDS and SD codes with the same underlying structure?

Key findings

• The proposed construction achieves maximal recoverability for all valid erasure patterns, including any $m$ failed disks and up to two additional erased sectors.
• The required field size for SD codes is bounded by the total number of array elements, ensuring practical feasibility.
• For PMDS codes, the field size requirement is at most quadratic in the total number of sectors, which is efficient for real-world deployment.
• The construction is proven correct via algebraic arguments showing invertibility of critical field expressions under all erasure scenarios.
• The method generalizes prior constructions for $(1;2)$ and $(2;2)$ codes, unifying them under a single framework.
• The approach supports both PMDS and SD codes using the same core structure, differing only in parity-check matrix design.