Back up for a minute – Backblaze HD reliability stats show oldies can be goodies

Cloud storage and backup provider Backblaze has released a comprehensive report detailing reliability statistics for the hard drives it operated during the whole of 2021, with an interesting finding on its older kit. The report leaves out boot drives (which are SSDs) and focuses on Backblaze's data drives, of which the firm …

[badflorist]

What is a "Drive Day"?

Looks kind of fishy when one drive is listed as 80 months old and has 323,000 "Drive Days", while another dive is much younger at 14 months old, but has 11,600,000 "Drive Days".

[Gene Cash]

Re: What is a "Drive Day"?

Well, that's not a single drive... I think that's the total operating days clocked by all the drives of that model, and the age is the average age of that model.

So if you had 5 drives of model XYZ running for a month, that'd be 150 drive-days.

That would mean they have a shit-ton (or ass-load, depending) of the one with 11.6M drive days, and indeed, it's listed as 38K drives of that model.

[Quando]

It would be interesting to see if they have the volume of data read/written to/from each drive, and if that impacts reliability.

[andy4blaze]

Andy from Backblaze here: We collect SMART stats data at the current time on a daily basis, but none of those stats record data volume as best as I can tell, at least for hard drives. Some of the SMART stats for SSDs look like they could do that, or at least imply they could, but we'd have to dig into those numbers to see if 1) they are actually reported, and 2) they are what we think they are, and 3) how consistent they are across the myriad of SSD manufacturers, so I wouldn't get my hopes up.

[Denarius]

usage stats

Andy, thanks for info. However, whatever controllers inside each Tome and run by OS should have some device usage data surely . Just using SMART seems a trifle trusting. Just asking as my own past limited experience in big arrays allowed individual device data or at least virtual device stats were extractable, eventually.

[Down not across]

Thank you for the statistics

Andy, I'd just like to thank you for the data you provide (and have done for quite some time) on the drives. I find it really interesting and useful.

[Joe Gurman]

Do the stats distinguish....

....between pre-HGST acquisition and post-acquisition WDC stats? Not so long ago, WD had a reputation for some dodgy drive models. Since purchasing HGST, they've done considerably better, word-of-mouth-wise.

[andy4blaze]

Re: Do the stats distinguish....

Andy Klein here from Backblaze. The stats do not distinguish between pre and post acquisition, except by model number of the HGST drives starting with HMS (HGST), to HUH (shortly after the acquisition) to WUH (WD branded drives). When and how much influence WD has on the HGST team over time and vice-versa is unknown. But I will agree the recent WUH model drives are pretty HGST-esque in their reliability.

[elsergiovolador]

Drives crazy

That reminds me when I had to decompress a couple of TB of small files from a spinning drive. ETA after few hours was showing over 3 months.

Got expedited a fresh 8TB nvme and had the files ready for processing in two hours...

[spireite]

Interestingly though, Seagate models also make up a chunk of the worst AFRs.

The bigger question is whether the AFRs are really meaningful in a real-world consumer context.

If there is a drive with an AFR of 2%, does that mean I have a 2% chance of a failure in my machine?

[eldakka]

The AFR is proportional to the sum of the importance of the data and the freshness of (if any) of backups.

That is, if the data is critical and you don't have any backups, the AFR approaches 100%.

If the data is 'nice to have' and you did a full backup an hour ago, AFR approaches 0%.

[julian.smith]

Statistics is hard

Your questions show your ignorance of statistics - one of these days you've got to get going.

[Anonymous Coward]

Re: Statistics is hard

"Your questions show your ignorance"

Now that's perceptive! Duh, questions are indeed triggered by ignorance in some specific matter, aren't they?...

But, most importantly, they are fueled by the commendable desire to fix this ignorance. They are not meant as a means for the self-righteous bullies to signal their blinding superiority.

People asking a honest question, no matter how stupid it might sound to you, should be respected, there are plenty of domains you yourself are the stupid ignoramus to be made fun of.

(I'm not the OP BTW)

[Empty1]

publishing data

Plus brownie points for publishing what many would say was commercially sensitive data.

Plus more brownie points for responding to questions publicly.