A new standard in fault tolerance for broadcast SANs

摘要

In the computer industry, the term 'crash' is defined as the sudden failure of a software application or hardware device. While a software crash usually creates an interruption in service, the dreaded hard disk crash tends to also cause the loss of data. Due to the mechanical nature of a high speed spinning disk, hard disk crashes are often permanent. Not surprisingly, disk manufacturers hedge against excessive drive failures by employing many internal data recovery mechanisms, including error correction, retries and sector remapping. Some errors that can result in seek, read and write failures are masked by successful retries. Only when a drive exceeds its retry count and runs out of sectors for remapping is it considered 'failed'.