Protecting RAID Arrays against Unexpectedly High Disk Failure Rates

Abstract

Disk failure rates vary so widely among different makes and models that designing storage solutions for the worst case scenario is a losing proposition. The approach we propose here is to design our storage solutions for the most probable case while incorporating in our design the option of adding extra redundancy when we find out that its disks are less reliable than expected. To illustrate our proposal, we show how to increase the reliability of existing two-dimensional disk arrays with n2 data elements and 2n parity elements by adding n additional parity elements that will mirror the contents of half the existing parity elements. Our approach offers the three advantages of being easy to deploy, not affecting the complexity of parity calculations, and providing a five-year reliability of 99.999 percent in the face of catastrophic levels of data loss where the array would lose up to a quarter of its storage capacity in a year.