Performance evaluation of the Mojette erasure code for fault-tolerant distributed hot data storage

TL;DR

This paper evaluates the Mojette erasure code, based on the Mojette transform, demonstrating superior performance over Reed-Solomon codes in fault-tolerant distributed hot data storage, enabling more efficient I/O operations.

Contribution

It introduces the Mojette erasure code and shows its performance advantages over traditional algebraic codes like Reed-Solomon in distributed storage systems.

Findings

Mojette code outperforms Reed-Solomon in coding and decoding speeds.

Performance gain up to 2x compared to ISA-L.

Suitable for hot data and I/O intensive applications.

Abstract

Packet erasure codes are today a real alternative to replication in fault tolerant distributed storage systems. In this paper, we propose the Mojette erasure code based on the Mojette transform, a formerly tomographic tool. The performance of coding and decoding are compared to the Reed-Solomon code implementations of the two open-source reference libraries namely ISA-L and Jerasure 2.0. Results clearly show better performances for our discrete geometric code compared to the classical algebraic approaches. A gain factor up to $2$ is measured in comparison with the ISA-L Intel . Those very good performances allow to deploy Mojette erasure code for hot data distributed storage and I/O intensive applications.