[Paper Review] More on Combinatorial Batch Codes

Sushmita Ruj, Bimal Roy|ArXiv.org|Sep 19, 2008

Advanced Data Storage Technologies1 references3 citations

TL;DR

This paper extends combinatorial batch codes to the general case of t > 1, providing exact and lower bounds for optimal storage N(n, k, m, t) and constructing column-uniform batch codes with rate 1/3. It resolves open questions from Paterson, Stinson, and Wei by deriving N(n, k, m) for n in the range (m choose k-2) ≤ n ≤ (k−1)(m choose k−2), and establishes new lower bounds for n₂(m, 2, k) using forbidden subgraph configurations in incidence matrix graphs.

ABSTRACT

Paterson, Stinson and Wei \cite{PSW} introduced Combinatorial batch codes, which are combinatorial description of Batch code. Batch codes were first presented by Ishai, Kushilevita, Ostrovsky and Sahai \cite{IKOS} in STOC'04. In this paper we answer some of the questions put forward by Paterson, Stinson and Wei and give some results for the general case $t>1$ which were not studied by the authors.

Motivation & Objective

• Extend the study of combinatorial batch codes from the t = 1 case to general t > 1, addressing an open problem posed by Paterson, Stinson, and Wei.
• Determine the optimal storage N(n, k, m, t) for n in the range (m choose k-2) ≤ n ≤ (k−1)(m choose k−2), filling a gap in prior work.
• Construct column-uniform combinatorial batch codes with fixed rate 1/3 for t > 1, improving storage efficiency.
• Establish new lower bounds for n₂(m, 2, k), the maximum number of items n for which a (n, 2n, k, m, 2)-CBC exists, using graph-theoretic constraints.
• Analyze the structural constraints of incidence matrices to prevent forbidden subconfigurations that violate batch code decoding requirements.

Proposed method

• Use the dual set system representation, mapping items to blocks and servers to points, with incidence matrices encoding storage and access patterns.
• Apply the transversal condition: for any k columns, a k×k submatrix with a full transversal (one 1 in each row and column) must exist to ensure decodability.
• Model the incidence matrix as a graph where columns with two 1s represent edges, and analyze forbidden subgraphs (e.g., triangles, squares, or 4-edge graphs on 4 vertices) that violate t = 2 decoding constraints.
• Derive bounds using extremal graph theory: the maximum number of edges in a triangle- and square-free graph on m vertices is greater than 2 + 3⌊(m−3)/2⌋.
• Construct augmented incidence matrices A′ by adding columns to the base matrix A of all 2-element subsets, ensuring no forbidden subgraphs are formed.
• Use combinatorial counting to determine the number of columns that can be safely added while maintaining the t = 2 decoding condition across all k-subsets of items.

Experimental results

Research questions

• RQ1 What is the optimal storage N(n, k, m, t) for t > 1 when n is in the range (m choose k-2) ≤ n ≤ (k−1)(m choose k-2)?
• RQ2 Can column-uniform combinatorial batch codes with rate 1/3 be constructed for t > 1?
• RQ3 What are the tightest lower bounds for n₂(m, 2, k), the maximum number of items n for which a (n, 2n, k, m, 2)-CBC exists?
• RQ4 How do forbidden subgraph configurations (e.g., triangles, squares, or 4-edge graphs on 4 vertices) constrain the construction of t = 2 batch codes?
• RQ5 What is the relationship between the number of items n and the number of servers m when t > 1, particularly for k > 2m?

Key findings

• For t > 1, the paper derives the exact value of N(n, k, m, t) when (m choose k-2) ≤ n ≤ (k−1)(m choose k-2), extending prior results that only covered n ≥ (k−1)(m choose k-1).
• The optimal storage for t = 2 and k = 9 is bounded below by n₂(m, 2, 9) ≥ (m choose 2) + 2⌊m/3⌋, based on avoiding 3-edge forbidden configurations in the incidence graph.
• For k = 10, the lower bound is n₂(m, 2, 10) ≥ (m choose 2) + 2 + 3⌊(m−3)/2⌋, derived from forbidding 4-edge graphs on 4 vertices and using extremal graph theory.
• The storage space required for t = 2 is approximately half that of t = 1, as shown in the comparison table: for m = 10, k = 5, N(t=1) = 640, N(t=2) = 360.
• The paper proves that if k > 2m, then n₂(m, 2, k) < (m choose 2), since k columns would span only m rows, violating the t = 2 constraint.
• The results generalize Corollary 3.3 and 3.5 of Paterson et al. [2] (which were for t = 1) to t > 1, showing that the new bounds are consistent with and extend the earlier work.