Server class disk drives: how reliable are they?

Abstract

Hard disk drive manufacturers frequently set high expectations for drive reliability from their specifications, test results and "global returns database". However, field reliability experienced by customers often does not match those expectations. Actual disk drive reliability may differ greatly from the manufacturer's specification and from customer to customer because of dependence on a variety of factors, some of which are completely independent of the drive design or manufacturing process. Thermal environment and duty cycle, inherent characteristics of the drive itself, architecture and logic of the system in which it is used, and the data collection and analysis process itself are all sources of significant variability. Together, these can create a range on mean time between failures (MTBF), if MTBF is indeed the correct metric, of 350,000 to 1,200,000 hours over the lifetime of a population of server class disk drives. This paper further elaborates on these four causes of variability and explains how each is responsible for a possible gap between expected and measured drive reliability.