[Paper Review] Malleable Coding with Fixed Reuse

Lav R. Varshney, Julius Kusuma+1 more|arXiv (Cornell University)|Sep 4, 2008

Advanced Data Storage Technologies24 references3 citations

TL;DR

This paper introduces a malleable coding framework that enables efficient reuse of compressed codeword prefixes when updating data, balancing compression efficiency and update cost. It establishes a single-letter characterization of the achievable rate region using an auxiliary random variable, revealing that high compression near entropy forces negligible reuse due to low common information between source versions.

ABSTRACT

In cloud computing, storage area networks, remote backup storage, and similar settings, stored data is modified with updates from new versions. Representing information and modifying the representation are both expensive. Therefore it is desirable for the data to not only be compressed but to also be easily modified during updates. A malleable coding scheme considers both compression efficiency and ease of alteration, promoting codeword reuse. We examine the trade-off between compression efficiency and malleability cost-the difficulty of synchronizing compressed versions-measured as the length of a reused prefix portion. Through a coding theorem, the region of achievable rates and malleability is expressed as a single-letter optimization. Relationships to common information problems are also described.

Motivation & Objective

• To address the trade-off between compression efficiency and malleability cost in data updates.
• To model scenarios where only a fixed prefix of a compressed codeword is reused during updates.
• To formalize the problem as a multiterminal information theory problem with a focus on codeword reuse.
• To derive a complete characterization of achievable rates using information-theoretic optimization.
• To establish connections between malleable coding and classical problems like Gács–Körner common information.

Proposed method

• Formalizes malleable coding as a three-terminal problem: original source, updated source, and a shared auxiliary variable for fixed prefix reuse.
• Uses a single-letter optimization involving an auxiliary random variable U to characterize the achievable rate region.
• Applies the implicit Markov property to simplify the analysis of the rate region under memoryless source assumptions.
• Establishes a coding theorem that expresses the trade-off between original rate, reused prefix rate, and new codeword rate.
• Leverages the data processing inequality and successive refinability to prove bounds on the rate region.
• Connects the problem to Gács–Körner common information to reveal fundamental limits on reuse when sources are highly correlated.

Experimental results

Research questions

• RQ1 What is the fundamental trade-off between compression efficiency and malleability cost when reusing a fixed prefix of a compressed codeword?
• RQ2 How does the common information between original and updated source sequences constrain the amount of codeword reuse?
• RQ3 Can the rate region for malleable coding be fully characterized using a single-letter information-theoretic expression?
• RQ4 What happens to the reuse fraction when both original and updated sources are required to be compressed near their entropy?
• RQ5 How does the structure of the auxiliary random variable relate to the achievable rates and malleability cost?

Key findings

• The achievable rate region for malleable coding is completely characterized by a single-letter optimization involving an auxiliary random variable.
• When the joint distribution of original and updated sources is indecomposable and rates are near entropy, the reused fraction asymptotically vanishes.
• A connection to Gács–Körner common information shows that low common information forces high malleability cost.
• The rate of the new codeword is bounded by the sum of the reused rate and the conditional entropy of the new source given the auxiliary variable.
• The upper bound on the new codeword rate difference relative to reuse rate difference is at most 1, implying a linear trade-off.
• The continuity of entropy in variational distance ensures asymptotic convergence of conditional entropies under lossless coding.