The most surprising finding to me (as already said in the article) is that even at BB's scale, the sample size (and time period) for lot of data categories are simply not statistically large enough to conclude much, at least not with this quarterly data. You can see it clearly when the failure rate fluctuates heavily QoQ and has lots of outliers when count of units is (relatively) low.
You still can get some trends, but for instance if you want to find say the "best" model, this probably isn't too helpful.
I guess it just shows that the failure of an HDD is a very random thing.
''The drive with the lowest AFR (0.28 percent) and at least 2.2 million drive days is Western Digital Corporation (WDC)'s 16TB WUH721816ALE6L4 (Backblaze has 14,098 units).''
''The model with the highest AFR (2.57 percent) and at least 2.2 million drive days is Seagate's 4TB ST4000DM000 (Backblaze had 18,070 units).''
You still can get some trends, but for instance if you want to find say the "best" model, this probably isn't too helpful.
I guess it just shows that the failure of an HDD is a very random thing.