Maximal Achievable Service Rates of Codes and Connections to Combinatorial Designs

TL;DR

This paper analyzes the service-rate region of distributed storage systems using linear codes, deriving bounds and conditions for optimal request rates, and connects these to combinatorial design theory for improved understanding.

Contribution

It introduces bounds on maximum request rates for linear codes in distributed storage and links these bounds to combinatorial design structures, providing exact limits for certain codes.

Findings

Bounds are tight for all non-systematic codes with known SRRs.

Upper bounds are achieved when supports of minimum-weight dual codewords form a 2-design.

Exact demand limits are determined for binary Hamming codes.

Abstract

We investigate the service-rate region (SRR) of distributed storage systems that employ linear codes. We focus on systems where each server stores one code symbol, and a user recovers a data symbol by accessing any of its recovery groups, subject to per-server capacity limits. The SRR--the convex polytope of simultaneously achievable request rates--captures system throughput and scalability. We first derive upper and lower bounds on the maximum request rate of each data object. These bounds hold for all linear codes and depend only on the number of parity checks orthogonal to a particular set of codeword coordinates associated with that object, i.e., the equations used in majority-logic decoding, and on code parameters. We then check the bound saturation for 1) all non-systematic codes whose SRRs are already known and 2) systematic codes. For the former, we prove the bounds are tight. For systematic codes, we show that the upper bound is achieved whenever the supports of minimum-weight dual codewords form a 2-design. As an application, we determine the exact per-object demand limits for binary Hamming codes. Our framework provides a new lens to address the SRR problem through combinatorial design theory.