On Some Universally Good Fractional Repetition Codes

TL;DR

This paper demonstrates that certain fractional repetition codes, constructed via partial regular graphs, ring, and t-constructions, are universally good for distributed storage systems, optimizing reliability and repair efficiency.

Contribution

It proves that codes built with partial regular graphs are universally good and introduces new universally good codes using ring and t-constructions.

Findings

Codes from partial regular graphs are universally good.

New universally good codes are constructed using ring and t-constructions.

These codes optimize repair and storage in distributed systems.

Abstract

Data storage in Distributed Storage Systems (DSSs) is a multidimensional optimization problem. Using network coding, one wants to provide reliability, scalability, security, reduced storage overhead, reduced bandwidth for repair and minimal disk I/O etc. in such systems. Regenerating codes have been used to optimize some of these parameters, where a file can be reconstructed by contacting any k nodes in the system and in case of node failure it can be repaired by using any d nodes in the system. This was further generalized to Fractional repetition (FR) codes (a smart replication of encoded packets) on n nodes which also provides optimized disk I/O and where a node failure can be repaired by contacting some specific set of nodes in the system. Several constructions of FR codes using graphs and combinatorial designs are known. In particular, some constructions of codes for heterogeneous DSSs are given using partial regular graph (where number of packets on each node is different) and ring construction. In this work, we show that the codes constructed using the partial regular graph are universally good code. Further, we found several universally good codes using ring construction and t-construction.