CaptainCretin :
If you look at the 2017 figures, a number Seagate drives that were prevalent on the market had poor reliability.
[...]
Even if you ignore the 29% failure rate of one Seagate drive type, the majority of drives with a higher than average failure rate were Seagate.
Kudos to you for at least trying to understand the statistics. Unfortunately, this is something easily messed up by people not well versed in statistics. The actual table you need to look at is the second one in this link.
https://www.backblaze.com/blog/hard-drive-stats-for-2017/
That gives you the annualized failure rate, and the high and low confidence interval. Have you ever seen a poll in a newspaper which says that 60% of people were in favor of something, +/-5%? The confidence interval is that +/- 5%. It means that if the statistics are such that if they ran the poll 100 times, then 95% of the time their result would have been between 55% and 65%. (I'm simplifying a bit, but this is the easiest way to think of it).
The Seagate drive with the 29% failure rate has a confidence interval ranging from 0% to 120%. That is, Backblaze experienced a 29% failure rate, but they ran that model for so little time that statistically they can't tell if the drives really are that unreliable, or if they just got unlucky. If they ran their year of operations again 100 times using the same number of that drive model, they might get a 0% failure rate, they might get a 100% failure rate. Which is just another way of saying "We don't know what the failure rate for this drive is."
Statistical certainty (shrinking the confidence interval) comes from having a large sample size. And they only had 1255 operating days with this drive model (which is one of the models OP was considering) - by far the smallest number of any drives - and experienced a single failure. That number is too small to draw any reliable conclusions about the model's failure rate. So you can basically ignore the 29% failure rate for that model. It might be right, it might not. Backblaze simply doesn't have enough data to draw any conclusions about that model either way. By all rights they shouldn't even have included it in the table.
Looking over the rest of the drives on that table, the next-worst performer is another Seagate model, but this time with a large enough sample that you
know that there's a 95% chance that its failure rate is
at least 10.6%. This is followed by two WD models with failure rates of
at least 4.5% and 3.9%, before you get to the next Seagate model with a 2.9%-3.1% failure rate.
If you look at the high end of the confidence interval, you find one Seagate model which
might be as bad as 21.5%. This is followed by two WD models which might be as bad as 5.9%, then another WD model which might be as bad as 5.8%, a Toshiba model which might be as bad as 5.7%, and a HGST model which might be as bad as 4.5%, before you get to the next Seagate model which might be as bad as 3.4%.
If you're trying to interpret their data to form a "avoid/buy company x" rule, their 2017 report is really damning for WD.
All three WD drives they used in 2017 had an unusually high failure rates; two of them with statistical certainty. In contrast, 4 of Seagate's models had low failure rates, 2 slightly high failure rates, 1 very high failure rate, and 1 there wasn't enough data to draw a conclusion. So if you were picking a brand (which I don't suggest you do) based on their 2017 report, then WD would be the worst choice. The majority of the drives with high failure rates were Seagates simply because Backblaze uses more Seagate drives than any other brand, not because Seagate is less reliable.
And my personal experience with drives have been two WD failures, one Seagate failure, one Samsung failure, one Toshiba failure, one IBM (which became HGST) failure, one Conner failure, one Fujitsu failure. Ironically, I have one Maxtor drive which I've been trying to kill for over a decade. Maxtor (before it was acquired by Seagate) was considered a low-quality discount brand. I bought a 300 GB Maxtor because of a killer sale price. It became the drive I tossed into my bag when I went to visit a client where I might need an extra HDD to copy data onto. It's been on dozens of car drives, jostled around in my bag (I took no precautions to protect it since I fully expected it to die), temporarily installed at 3 clients' computers for an extended period, I've even dropped it onto carpet a couple times. This so-called low-quality drive just doesn't want to die.
Which brings me to my point. Unless you're running drives for a cumulative hundreds of thousands or millions of days like Backblaze is, your personal experience with drives is statistically meaningless. An individual simply doesn't accrue enough operating time with drives to generate statistically reliable conclusions. The drive you got might've been bad, or you might've gotten unlucky. The only statistically reliable stats I've seen on HDD failure rates are Backblaze's reports, and the now-defunct Storage Review HDD reliability database.
Facebook
Twitter
Instagram