Why Backblaze Hard Drive Reliability Data is BS

The hard drive reliability data that Backblaze releases on a regular basis is used by many people to make purchasing decisions. We show how the data is flawed due to inappropriate use of the wrong hard drive models and conflating of data which mis-represents manufacturer reliability.

[Ludenson]

Nice video, but bad title!
The data are not BS, but they are often used wrong.
You used the data in the video to prove a point - how can you then call the data BS???

[sf]

Nice explanation and you make a good point not to compare different drive classes against one another, and to select a drive properly for class of use.

What you don't mention is that the table is VERY valid for comparing similar drive classes. This table clearly show (although the population of different models is smaller than I would like) what I've experienced empirically that Seagate drives are much less reliable than the HGST drives (AFAIK even though they are desktop they are rated 24/7).

Also note that the HGST "Desktop"" drives appear to be holding up much better than the WDC "1-8 Bay NAS" drives.

I would suspect that in the Backblaze use case that the drives would not really be doing "enterprise" class service. Given they are archive storage, the drive would fill up, and then do very little work unless it was called upon for restoring a file. This would likely be less onerous than many desktops - especially a windows system drive that is constantly being hammered with all sorts of crap that windows generates internally.

So while I appreciate the good points that you have made, I think calling the Backblaze Hard Drive Reliability Data BS does nobody any service.

[_droid]

How is the data "BS"? It's accurate data as far as I know and you provide no proof it isn't. Are you saying they publish false information?

If people can't figure out what the difference is between models that's their own fault.

[matth]

First, I believe in the latest report Backblaze was suggesting that most of the enterprise drives are new so unsurprisingly they have a lower failure rate.

Also asserting the data is wrong seems a bit silly. I think a better assertion would be people can easily misinterpret the stats but having the data available is pretty handy. It allows people to have discussions based on evidence rather than the anecdotal evidence that fills the industry. Like in this case where you make an arguement for enterprise drives.

Finally I don't believe Backblaze ever said enterprise drives were the same, they just didn't think the additional value offered justified the increased price. I agree they haven't shared their costing model but neither has anyone else including you since calculating the true cost of a failed hard drive is so variable based on the storage system and replacement process surrounding it.

[RustyTheGeek]

Backblaze Data is what it is... collected statistical data. I didn't get the impression BackBlaze had an agenda other than to help drive people to their site to maybe sell data space. They give different summary views of the data for convenience to help in its analysis and they share the raw numbers as well.

Like any statistical data, it can be interpreted many ways. It's up to the user to figure that out for themselves. I agree with the comment below, compare the drives with some class discrimination and get the results you need.

Are Enterprise Drives more reliable? Sure! (I would certainly hope so!) The question each user must answer is what is that reliability worth and can they tolerate a higher failure rate for a lower cost. The data BackBlaze provides helps in making that decision. I don't think it's a blanket justification for buying Desktop Drives for Enterprise Use. It just helps one determine the amount of risk they can tolerate with a given drive model.

I for one am very grateful that we have several years worth of reliable data for so many various drive models to help avoid a potentially bad choice (times x number of drives that match).

[ernie]

Nice work . I must say BB operates at a different scale at many levels. That they share the data is intetesting and kinda cool. Who else does that? People watching and not operating at or near that scale may just take a step back and simply enjoy the info-tainment for what it is . :)

[wildcats]

imho they did their math on this, if a drive fails the data is still preserved cause of redundancy which means data backup service will reliable. sure cheap hard drive will most likely to fail than enterprise but it should still give better roi specially on the scale that they are operating. Of course if you create a data rack with smaller scale to store data for your company the storage is considered as a tool and not a money making source enterprise disk will be better. But if I will also try to create a cloud storage service then their advise is fine as I can make the cost lower for consumer and as long as I can manage the data security and drives will give roi before they fail.

[elvisjonesus]

I have a tight nas budget. Refurb enterprise drive or new desktop drive ?

[charliebrownau]

Gday Whats your view on backing up using BR-R or DVDR 8.5 gig discs ?

[cataria]

my god u just added individual drive failure rates together and divided, completely ignoring drive days/drive.
2 of the enterprise drives having sub 20k drive days, yet u claim, that backblaze's data is bs, while u screw up the simpler math?
furthermore the 2 enterprise drives with tons of days are some of the most reliable drives i have seen in all of backblaze's stats i looked at thus far, the 4 TB hgst megascale drives.
so are we looking at 2 amazing models and 2 fresh sets of 8 TB drives with less than 20k drive days each creating the artificial and misleading look of 0.23% for enterprise drives? yes, yes absolutely.

if u're interested the st8000nm0055 ended up being at 0.88% AFR for 2018 and huh728080ale600 ended up being at 0.68% AFR 2018, still decent, but definitely not the 0.00%, that u used at ONLY 1560 drive days for the seagate

[redstang]

He pretty much comes to the opposite conclusions as of when that video was made and also makes the point that the consumer drives are so much cheaper that they don't care because the failure rates still are not high enough to make a difference.

[duncanhaskell]

I suppose that beacsue the I in RAID stands for Inexpensive so I can understand why they are experimenting with cheap drives, but yes the stats are not written up with a high level view of what the data actually means, which is going to confuse people.

[gskibum]

Are you the hand actor that plays the Swedish Chef?

[HikikomoriDev]

My issue wasn't even the data realibility chunk but how slow the speeds where to just get to upload... I had some 1.5TB of storage to clone and it never got completed by the 15 day trial period... You can notice the speed difference when you upload some large video file to YouTube and notice the speed difference... Backblace took ages :(

[kian]

I've only used consumer drives and most of them are Seagate, among roughly the 30-40 drives I've had, Seagate, WD and HGST never died on me no matter what, Samsung and Toshiba both did.

[bogdan]

Very nice explanation Casey.Also for my laptops i use ssds because they fail less from mechanical shock.

[mdd]

RAID 5 has been considered a 'no-no' with spinning drives for years, already....

[aussieexpat]

Great points. I thin the choice still comes down to wether the cost of replacing broken drives is significant.

Even if you get a NAS drive for a NAS, the failure rate is not zero. So it comes down to how much that drive replacement activity is worth to you.

[roudroabaabil]

Their DATA is not a BS, their demonstration may be. It depends on the way of using that data. If they had any intention, then they would hide the model numbers. :)

[edwardkong]

Regardless of any stats I would never buy Seagate drives ever again. Had 5 drives (including an OEM drive in a Sony dvr) over 10 years and all failed within a year after warranty.