Online backup provider Backblaze has a new batch of reliability data from its army of almost 40,000 mechanical drives. This time around, the company is delving into SMART statistics—specifically, which ones are good predictors of impending doom.
As it turns out, only a handful of attributes correlate with drive mortality. Most do not, and the fact that "different disk drive manufacturers report SMART stats differently" complicates the matter. There are differences not only in how certain attributes are reported, but also in which ones are even present. We've noticed similar inconsistencies with the solid-state drives in our SSD Endurance Experiment.
Backblaze says its efforts to anticipate drive failures concentrate on just five attributes:
SMART 5 – Reallocated_Sector_Count
SMART 187 – Reported_Uncorrectable_Errors
SMART 188 – Command_Timeout
SMART 197 – Current_Pending_Sector_Count
SMART 198 – Offline_Uncorrectable
Here's how reported uncorrectable errors correlate with the annual failure rate:
The other correlation that really stands out is the reallocated sector count. As one might expect, drives with more bad sectors have higher failure rates.
If you want to do more digging, all of the SMART stats are available here. There aren't any breakdowns by drive model or manufacturer, but Backblaze does list which drives report which variables.