Which hard drive? 160,000 years experience analysed

Even if you’ve an office or two full of Macs, your experience of hard drives will never come close to that of large datacentres. We should therefore take careful note every quarter when Backblaze, a major provider of cloud services, gives us insight into the performance of its well over 70,000 drives. You can read Andy Klein’s review of that experience here.

Analyses from data centres like Backblaze’s have already taught us important lessons about selecting and using hard drives. Among those are:

  • Hard drives with the same manufacturing batch number tend to fail at about the same time.
  • Variation in average life for most drives is greater between batches of the same model, than between different models.
  • S.M.A.R.T. indicators are useful for predicting imminent drive failure, but otherwise of very limited value.
  • Some S.M.A.R.T. indicators are good indicators of imminent drive failure, but many indicators are of little or no predictive value.
  • Drives which run hot tend to die early.

In the last quarter of 2016, Backblaze used significant numbers of models from four different manufacturers (HGST, Western Digital, Seagate, and Toshiba), ranging in capacity from 3-8 TB. Most of their drives were made by HGST or Seagate. (HGST is the successor to Hitachi, which took on the ill-fated IBM Deskstar range and made a success of them. It is now owned by Western Digital.) Backblaze has generally avoided paying the additional cost of Enterprise-class drives, so most of the disks mentioned here are consumer-class.

Annualised failure rates across all drives average at just under 2%, with HGST generally rather better than that, and Seagate slightly worse. However, HGST’s worst-performing model, a 3 TB drive, was significantly worse (3.6% annualised failure rate) than Seagate’s best (1.7%). Among the more used drives, HGST 4 TB (models HMS5C4040ALE640 and HMS5C4040BLE640) were excellent, with 0.5-0.6% annualised failure rates. However Seagate 4 TB drives (models ST4000DX000 and ST4000DM000) were among the poorer performers, with annualised failure rates of 14.7% and 2.7%.

Looked at across the whole year, the picture is very similar. This was not a ‘rogue’ quarter. Good performers included HGST 3 TB models (HDS5C3030ALA), all three HGST 4 TB models, and Seagate 6 and 8 TB models.

Even more interesting are the pooled figures for the period April 2013 to December 2016, which covers drives ranging in capacity from 1 to 8 TB. Considering only those models for which there are data for more than a million drive days, annualised failure rates ranged from 0.5% (HGST 4 TB) to 26.7% (Seagate 3 TB). Few HGST models performed worse than average, but few Seagate models were better than average. The general trend in larger (4 TB and above) drives is to lower failure rates, not higher. Of the different capacities used over the years, highest failure rates have generally occurred in 1.5 and 3 TB models.

At the moment, if you’re looking to buy hard disks, HGST’s 4 TB models seem to have excellent track records and low annual failure rates for their cost. If you’re buying more than one, ensure that they come from different batches. If you have the choice, and can afford to pay the extra, try to get drives with warranty periods longer than three years. Buy carefully, keep your drives cool, keep them spinning all the time, and they should last you a good 4-5 years or more.

Backblaze’s experience is now enormous: these data cover a total of almost 160,000 years of spinning hard drives. We all should be very grateful to them for sharing it.