A model of Erasure-Coding-Based Distributed File System

Abstract

Today's exponential growth in network bandwidth and storage capacity has inspired different classes of distributed storage infrastructures. However replication has become the actual way for storing data across data centers, no matter the system is centralized framework or peer-to-peer infrastructure. Erasure code, which is well researched and more space efficient than replication, has multiple tradeoffs that need to be seriously considered. In this paper, we present an erasure-coding-based model coupled with hadoop library, and we call it the Erasure-Coding-Based Distributed File System. We discuss the detailed design of our model and provide some recommendations on other coding strategy for distributed file systems.