Lifetime assessment and failure rate characterization of hard disk drives using large-scale data

As an indispensable modern electronic device, the hard disk drive (HDD) has been hyped with life as long as 2.5 million h, equivalent to about 285 years. However, drive failure occurs in our everyday use, casting big doubts on these claims. By analyzing a large amount of drive failure data, covering 34,000 disks with 160 various drive models over a 10-year time span, this paper assesses the real field life of the two major types of storage drive: HDD and solid-state device (SSD). A general failure rate function, named polynomial failure rate, is proposed to capture the characteristic of the drive failure over time. The observed failure rate finds diverse types of shape other than the conventional thought of bathtub. For life duration assessment, traditional Exponential and Weibull analysis are performed along with the methods by using the proposed polynomial failure rate function. The best life model is selected by using likelihood ratio test or G-Test. By using the large amount of field data, the life of the hard drive is compared between their models and disk types, and it also compares with the life specified in the datasheet. Moreover, a discrete yearly failure rate, as alternative to MTBF (Mean Time Between Failure), is proposed to gauge the drive life that finds more reasonable for nonconstant failure rate.