Analysis of Repair Cost in Distributed Storage Systems with Fault-Tolerant Coding Strategies

摘要

To achieve reliability in distributed storage systems, fault tolerance techniques like replication strategy are adopted. As the rapid growth of data, distributed storage systems have been transitioning replication strategy to coding strategies like Reed Solomon codes to achieve higher storage efficiency. But the repair cost of Reed Solomon codes in terms of network bandwidth is high. For repair efficiency, a new class of codes called Regenerating Codes are proposed and become more popular. However, how to quantify and evaluate the repair cost of these coding strategies at the system level remains unexplored. In this paper, we propose a metric of the repair cost at the level of whole systems, and then compare the two main classes of codes Reed Solomon codes and Regenerating codes. Our goal is to provide system designers with evaluation methods of the system level repair cost. Thus, system designers can choose optimal coding strategies according to their certain systems.