Backblaze has released its data set for the full year 2018 and there are some surprises in it this time around. Annual failure rates for drives have fallen sharply in comparison to previous years as smaller capacity HDDs have been replaced by higher capacity counterparts. This isn’t automatically what you’d expect.
The Annualized Failure Rate (AFR) has fallen every year since 2016, from 1.95 percent to 1.25 percent. That’s a significant decline. Importantly, while this trend doesn’t hold true for every drive family, there are multiple HDDs that show lower AFRs in 2018 than in 2016. Seagate’s 10TB drive family has been particularly good, with a 0.33 percent AFR for 2018, improving on the already-low 0.89 percent AFR for 2017.
For those of you wanting an AFR comparison that stretches back farther than 2016, Backblaze has provided that data as well. The following is the AFR rating for HDDs from 4/20/2013 through 12/31/2018 for all drives still active as of 12/31/2018.
HGST drives continue to impress as far as overall AFR rates and this isn’t the first year that’s been true. Overall, these are some of the most reliable drives over time that Backblaze has tested. The complete AFR rate over all five years is 1.68 percent — higher than the 2018 data alone, but still within a very reasonable window.
I’ve written a version of this for every Backblaze post we do, so here’s the latest. Backblaze’s data set is not perfect. The company uses consumer drives in a decidedly enterprise context, which doesn’t remotely reflect the usage patterns these drives would receive if they were being used by an ordinary customer. It constructs and maintains its own storage pods and has iterated on these designs over time, often with improvements intended to reduce vibration or noise. This could also theoretically have an impact on drive reliability. As the video below illustrates, hard drives don’t like it when you scream at them. Always make certain to communicate with your spinning media in a thoughtful, calm voice.
The reason we continue to discuss and feature Backblaze’s data set is the same reason we refer to the Steam Hardware Survey. We know the data sets are imperfect — the SHS’s list of video cards doesn’t include multiple AMD GPUs like Vega 56 or 64 and only added the RX 570 last month. The total number of “Other” GPUs listed is 10.92 percent — more than enough to meaningfully impact the AMD versus Nvidia split listed at the top level of the page. But while these data sets are imperfect, they’re also the best window we have into an important question. HDD manufacturers do not release the kind of reliability information that HDD consumers often want and the handful of third-party studies on the topic frequently don’t identify drive vendors or models.