A total of 1,225 drives failed in 2016, which means the drive failure rate for 2016 was just 1.95 percent, a improving over the 2.47 percent that died in 2015 and miles below the 6.39 percent that hit the garbage bin in 2014.

What, did they start using actual server drives in their enclosures, or something?

Think People here would be more interested in SSD failure Rate's
By size and Brand
After all its 2017 and Spinning rust is on its way out !!! :)

dorsetknobAfter all its 2017 and Spinning rust is on its way out !!! :)

Sadly it's not in the datacenter, though it won't be long.

R-T-BWhat, did they start using actual server drives in their enclosures, or something?

Last I checked they still used junk drives lol

cdawallLast I checked they still used junk drives lol

HGST doesn't even have a 2016 3.5" lineup that isn't enterprise I don't think... maybe they are buying old stock or using travelstar 2.5", lol.

R-T-BHGST doesn't even have a 2016 3.5" lineup that isn't enterprise I don't think... maybe they are buying old stock or using travelstar 2.5", lol.

They don't swap drives out yearly do they? I thought it was carry over minus replacement.

Backblaze telling us what's the most reliable HDD brand by using desktop-grade HDDs in server racks.

People just can't get that through their thick skulls for a few years now.

cdawallThey don't swap drives out yearly do they? I thought it was carry over minus replacement.

Oh, duh. That would make sense yeah.

Dj-ElectriCBackblaze telling us what's the most reliable HDD brand by using desktop-grade HDDs in server racks.

People just can't get that through their thick skulls for a few years now.

As someone who maintains few servers and knows web hosting market, I know why. They have to use cheap drives in order to compete in the market.

alucasaAs someone who maintains few servers and knows web hosting market, I know why. They have to use cheap drives in order to compete in the market.

I don't think that was his point. The point is these drive behave differently in server environments than in desktops. So these numbers don't really tell the reliability of the drive in the desktop environment.

newtekie1I don't think that was his point. The point is these drive behave differently in server environments than I'm desktops. So these numbers don't really tell the reliability of the drive in the desktop environment.

Although if a metric shit ton of drives all fail at the same time, I would find that information to be useful as a steer clean notice.

Exactly and this best demonstrates that in the long term:


And the short term:


Seagate ST4000DX000 has a scary-high failure rate at 13.57%. Good idea to avoid buying those.

The fact that only 9% of their Seagate ST1500DL003 drives are still operational is also scary. If I had one of those, I'd be getting the data off of it and disposing of it.

That said, the take away from Backblaze is pretty simple: of the 85,467 hard drives they purchased, 5,380 had failed. That amounts to 6.3% of drives purchased since April 2013. Failure rates have been declining from 6.39% in 2014, to 2.47% in 2015, to 1.95% in 2016. So not only are hard drives continuing to prove they're generally reliable, they're also getting more reliable with increasing densities. This is very good news! :D

....For all the criticism about not using actual server drives, these guys likely use some form of open stack - swift. The I/O loads are very very different than a traditional raid setups.

bpgt64....For all the criticism about not using actual server drives, these guys likely use some form of open stack - swift. The I/O loads are very very different than a traditional raid setups.

It's not really loads that are the concerns, but endurance.

R-T-BIt's not really loads that are the concerns, but endurance.

Endurance really isn't a concern the drives are used alot less than a traditional raid. They burn out alot faster, but they have 5-6 times as many copies of a block of data across multiple drives/clusters.

Ahh Backblaze's statistics, the ultimate reference for anyone wanting to know which HDDs take the longest on average to die when you put 50 of them in an enclosure the size of 50 HDDs with zero cooling ^^

bpgt64Endurance really isn't a concern the drives are used alot less than a traditional raid. They burn out alot faster, but they have 5-6 times as many copies of a block of data across multiple drives/clusters.

I'm referring to ontime (24/7) and cooling conditions (hot as hades) specifically.

Even with the server/desktop environments contrast, I might accept their results for Hitachi and Seagate as worst case scenario figures, but the sample sizes for Toshiba and WD is too low for them to be of any use!

I've been using BackBlaze for over a year now because I have "$5 data" to back up; this is not for my fancy data :)

UbersonicAhh Backblaze's statistics, the ultimate reference for anyone wanting to know which HDDs take the longest on average to die when you put 50 of them in an enclosure the size of 50 HDDs with zero cooling ^^

Thank you! I know you said this tongue-in-cheek (they have fans) but it has been nice to see this environment play out in such detail over the last 7 years.

A bit anecdotal, but the HGST in my home server is purring away while my Seagate just pasted it warranty date... and bad bits started to appear ;)

Here's some food for thought, if you stack 15 HDDs on top of each other, then surround them by more stacks of 15 HDDs then not only do the ones nearer the centre of each stack run much hotter but the stack in the middle also runs much hotter.

Considering that operating temps will vary so massively from drive to drive it basically invalidates any attempt to log the failures rates as although all the drives are being abused there are many which are being abused more and no common methodology at work at all.

I.E the two drives on the list with the highest failure are Seagate DX drives (the newest ones on the list), now if they bought a load in bulk and started using them to replace failed drives then they would be putting most of them in the areas where drive failure is most likely to occur, thus hyper inflating their failure rates.

Another point is that if you have 1000 DriveX and 100 DriveY and break 50 of each then the failure percentage will be vastly different even though you killed the same amount of drives.

cdawallAlthough if a metric shit ton of drives all fail at the same time, I would find that information to be useful as a steer clean notice.

Yeah, but even BackBlaze admits that when that happens, they just don't include it in the results. Like when a certain WD model had a 100% failure rate, they said they just left it out of the data because it would "throw of the numbers." Yeah, that isn't how you present accurate data.

Also, some of their number seem very fishy to me. I didn't realize it until a read Ford's post.

He took the 85,467 drives, and the 5,380 failures, and calculated a 6.3% failure rate. But BackBlace only claims a 3.36% failure rate.

So I started looking at some spot data for a few of the drives. Look at the Seagate ST1500DL003. They say they only used 51 drives, but somehow 77 failed? And that amounts to a 90% failure rate? I call BS. Besides the fact that it seems they had more drives fail than they actually had, the failure rate is wrong. And it isn't just that drive. Look at the Hitachi HDS723030ALA. They had 1,027 drives and 69 fail. That is just over a 6.7% failure rate, but they claim 1.92% Then look at the Seagate ST8000DM002. They have 8,660 drives and 48 failures. That amounts to a failure rate of 0.55%, but they claim 1.63%.

There is something screwy with their numbers.

newtekie1There is something screwy with their numbers.

Hopefully the percentages weren't calculated by the same processors as their data parity ^^

newtekie1There is something screwy with their numbers.

I bet that it is number of drive bays for that model HDD.

cdawallI bet that it is number of drive bays for that model HDD.

I don't think so, in the article text they say:

You’ll also notice that we have used a total of 85,467 hard drives.

So that column is the number of drives, not the number of bays.

Then whomever runs their excel document sucks at fomulas...