Western Digital, HGST top the list of most reliable hard drives

Hard drive failure data from cloud backup service Backblaze has Western Digital and Hitachi coming out on top -- and Seagate upping its game

Last year, cloud backup service Backblaze crunched statistics about which makes and models of the tens of thousands of drives humming away in its data centers held up best under stress. Hitachi and Western Digital came out at the top; Seagate, not so much.

Now Backblaze is back with another year's worth of stats, harvested from the consumer-level drives running in its custom-designed and open-sourced Storage Pod drive racks. The results, assembled from a data set more than twice as large as the previous year's, square with the earlier findings.

Hitachi (now HGST, a subsidiary of Western Digital) has the lowest failure rates across the makes and models surveyed. Western Digital itself came in second, with numbers only slightly less impressive than HGST's. "It’s hard to beat the current crop of 4TB drives from HGST and Seagate," Backblaze said in its blog post.

hard drive failure by manufacturer Backblaze

Hard drive failures by manufacturer over the course of one year. HGST, a Western Digital subsidiary, did best, but stats on WD's 6TB drive line remain preliminary.

Seagate, on the other hand, is another story. Its drives didn't do well in the first roundup and this year sported failure rates as high as 43 percent annually. As with last year, its 4TB models were far more durable than its other offerings, failing at around half the rate of the previous year.

What constitutes a failure to Backblaze? Aside from obvious mechanical problems -- the drive won't spin up or be recognized by the OS -- Backblaze included any drives that would not sync properly with a RAID array or reported SMART statistics that were out of the acceptable range. This last criterion can be tricky; Backblaze itself notes that SMART stat reporting isn't consistent between many drives. That said, the company believes a handful of the most critical criteria, such as the uncorrectable error count or the count of reallocated sectors, are reliable indicators of failure based on what it's seen in its drive pools.

The best results were with 4TB drives, which showed a marked decline in failure rates since the previous year's statistics -- both between HGST and Seagate. However, 3TB drive were less impressive, and Backblaze promised to dig into the story behind Seagate's striking failure rates there in a future post. Western Digital had no 4TB drives in the running, but Backblaze used 6TB drives from the company's line, the Western Digital Red. Its failure stats were less than 5 percent for the course of the year, but Backblaze cautioned it hasn't been using them for long enough to compute robust failure statistics.

Western Digital acquired Hitachi's hard drive business and turned it into HGST back in 2012; it was originally created in 2003 when IBM and Hitachi merged their hard disk manufacturing concerns. The HGST drives profiled in Backblaze's analysis were all Deskstar or Megascale models, the latter composed of 4TB drives designed for "low application workloads that operate within 180TB per year." Other drives in HGST's lineup include helium-filled 8TB and 10TB drives, with the helium providing greater capacity and lower power consumption, although Backblaze hasn't used those drives in its tests, preferring instead to stick with low-cost consumer drives purchased in bulk.

Backblaze has been using its data center as a source of eye-opening and sometimes hotly contested insights. Not long after its 2014 hard-drive reliability report, the company analyzed the effect of cooling on drive lifetimes. It found that keeping a drive cooler than its recommended operating temperature had no discernible effect on its longevity. Not everyone agreed with the conclusions, but few could find fault with Backblaze's underlying mission.

For those who want to crunch the numbers themselves, Backblaze plans to make available the raw data from the 2014 drive pool study in the next couple of weeks, along with more details on how it computed failure rates.