The disaster recovery artists over at Backblaze have published the latest of edition of the company's eye-opening hard drive reliability reports. Andy Klein, the report's author, says that Backblaze has been collecting data since 2013 and has tabulated 88 million statistical records pertaining to 93,000 hard drives. That data set measures a staggering 23 GB.
Backblaze's latest report refers to data collected during the fourth quarter of 2017. The annualized failure rates are an important metric, but they require careful reading—after all, the first chart on the report suggests that Seagate's ST4000DM005 drive is quite troublesome, but that particular drive had one failure among only 60 units in service for a combined 1255 days. The company doesn't report data from drive models with fewer than 45 units in its storage pool, so the sample size for the particular Seagate drive is near the company's minimum for inclusion in the results.
Over 2017, the company put 19,000 new drives to work while increasing the average size of a drive in service. Backblaze installed about 82 drives per day last year (including drive migrations and replacements) and replaced four failed units in an average day in the same year. Storage capacity jumped by 211 PB over the course of 2017. The company has tracked almost 117,000 drives move through its pool since April 2013.
Analysis of new and old data shows that the failure rate of 6 TB drives has fallen over the last three years. Klein cautions against drawing similar conclusions about 3 TB drives because of their shrinking numbers in service at Backblaze. A pair of Toshiba models had zero failures over the course of a year, though the company only has 45 units of each type. The company does have enough HGST drives in service to stand behind their failure rates of less than one percent.
Leon Chan over at Tom's Hardware analyzed the results of Seagate's ST8000NM0055 enterprise 8 TB model and compared them to the results from a consumer model of the same capacity. Over the time the drives have been in operation, the consumer devices actually have a slightly lower failure rate than the pricier enterprise versions. One should be careful not to draw conclusions about the overall enterprise hard drive market based on this single drive model, but it is interesting stuff nonetheless.
The most relentless gerbils can head over to Backblaze's reliability data download page and analyze the raw data themselves. The mildly curious can read Klein's summary here. Backblaze remains the only transparent source of so much information regarding hard drive reliability, though we should note that its usage patterns aren't a perfect mirror of the way a drive is used in the average desktop PC.