SoylentNews is people
SoylentNews
from the spinning-rust-still-has-its-place dept.
The cloud storage company BACKBLAZE has published another in their series of quarterly articles looking into Disk Drive failure rates.
The company had 68,813 spinning hard drives in operation. For Q3 2016 they have 67,642 drives, which is 1,171 fewer than their last quarterly report. The decline is because they have been migrating from their 2 terabyte (TB) drives to 8 TB models. They currently run a mix of 2, 3, 4, 5, 6 and 8 TB drives in their cloud storage system from a mix of different vendors.
The 8 TB drives are too new to reflect anything other than infant mortality rates, but all of the other sizes have been heavily used for years, such that some brand-specific trends are starting to appear.
The results are summarized in a table with the key metric being Annualized Failure Rate which is computed as follows: ((Failures)/(Drive Days/365)) * 100.
The Seagate 8 TB drives are doing very well. Their annualized failure rate compares favorably to the HGST 2 TB hard drives. With the average age of the HGST drives being 66 months, their failure rate was likely to rise, simply because of normal wear and tear. The average age of the Seagate 8 TB hard drives is just 3 months, but their 1.6% failure rate during the first few months bodes well for a continued low failure rate going forward.
Still, when you look at all the brands and models involved, the HGST brand seem to show the lowest failure rates historically.
With some reporting failure rates over 10% annually, mirrored drives may still be a wise choice for not trusting in the cloud.
(Score: 5, Insightful) by RedBear on Friday November 18 2016, @07:04AM
I can't wait to see a bunch of armchair statisticians once again explaining how having a data set of nearly 70,000 drives of various ages, with mixed sizes and brands, is totally useless for determining which drive brands and models may be more (or less) reliable than others. I still can't understand exactly what other source of such information people think would be better, but every time the BackBlaze drive reliability report comes up it seems to be met with scorn and derision. "You can't apply their data to inform your decision on what drives to buy for a home file server, that's a completely different use case!" they say. Whereupon I say, "Huh? But I'm... storing files on them... and I don't want to spend my money on expensive drives that are more likely to fail..." Insert confused face here.
Then again, I'm only a bear, and bears are said to have remarkably small brains.
How do YOU determine which brands and models of hard drives to buy for important file servers?
¯\_ʕ◔.◔ʔ_/¯ LOL. I dunno. I'm just a bear.
... Peace out. Got bear stuff to do. 彡ʕ⌐■.■ʔ
(Score: 1, Informative) by Anonymous Coward on Friday November 18 2016, @07:40AM
Well, since you mentioned it.
The standard critique is that backblaze puts consumer drives in an industrial environment because they buy the cheapest available models. That the drives are not designed for the amount of 24x7 vibration that occurs in a case with 48+ total spinning drives packed together. So failure rates don't match what home users will likely see and to support that argument they cite MTBF numbers for industrial spec drives versus consumer grade drives as being at least 10x better. So drives with good numbers will probably have good numbers in consumer environments, but drives with bad numbers may have much better numbers when used in a consumer-grade environment.
(Score: 2) by Jerry Smith on Friday November 18 2016, @10:00AM
For example the (former) WD Green vs WD Red.
All those moments will be lost in time, like tears in rain. Time to die.
(Score: 2) by MrNemesis on Friday November 18 2016, @12:34PM
Anecdata time...
Personally we've experienced higher failure rates in our NAS racks with the WD reds than we have with the WD greens (with their idle timer disabled via idle3-tools). Only have experience from the 3TB and 6TB ranges of both to compare.
Mechanically, the reds and greens are identical, it's only the firmware that differs as far as we can tell. Once the greens have the idle time disabled they're also identical in power usage and performance. The difference between the two is mostly just artificial market segmentation.
"To paraphrase Nietzsche, I have looked into the abyss and been sick in it."
(Score: 2) by bzipitidoo on Friday November 18 2016, @01:45PM
Speaking of anecdata, of the 4 drives I have purchased in the last 10 years for personal use, 2 were WD, and they both died early. I bought a WD black to replace the WD green when it died after just 9 months, then the black failed 4 years later and I decided I definitely didn't want another WD, and got a Toshiba to replace that. (The 4th drive is whatever 32G SSD is inside an Asus Vivostick, it's less than a year old and works fine.) I see that the Backblaze data shows a much higher failure rate for WD.
In the near 30 years I've been using hard drives, the only failures I've had are those 2 WD drives. I still have most of my old drives and they all still work.
(Score: 2) by RedGreen on Friday November 18 2016, @03:23PM
"In the near 30 years I've been using hard drives, the only failures I've had are those 2 WD drives. I still have most of my old drives and they all still work."
Well in my thirty + years of using them just about every seagate I have owned in the 1tb and above size have died still have one 1.5 and 1 1tb living out of ~20 purchased. The WD drives bought to replace them failures all of the green variety 3 have went titsup.
"I modded down, down, down, and the flames went higher." -- Sven Olsen
(Score: 2) by frojack on Friday November 18 2016, @10:12PM
Wait, what?
First you said
we've experienced higher failure rates in our NAS racks with the WD reds than we have with the WD greens
Then you followed up with:
Its only the firmware that differes ... Once the greens have the idle time disabled they're also identical in power usage and performance.
Your own data seems to suggest there is more difference than just firmware if your high priced drives are more problematic than your low priced drives.
No, you are mistaken. I've always had this sig.
(Score: 0) by Anonymous Coward on Friday November 18 2016, @03:02PM
IIRC they use datacenter drives for the hitachi drives. so the standard whining must be from wd/wd users. maybe their commercial drives are too expensive? either way i don't give a shit. i buy the hitachi.
(Score: 4, Informative) by frojack on Friday November 18 2016, @10:08PM
That the drives are not designed for the amount of 24x7 vibration that occurs in a case with 48+ total spinning drives packed together.
Having 48 drives spinning cause no vibration (none that wouldn't be completely absorbed by the drive's own case.
48 drives seeking might cause some vibration, but with 48 they are as likely to cancel each other out as to amplify each other.
I suspect this reasoning is EXACTLY the type of nonsense that RedBear was talking about. Silly un-provable claims made by arm chair wizards who run exactly 1 drive in their laptop.
With the average age of the HGST drives being 66 months, (5.5 years) and still showing the lowest failure rate, with over 4200 such drives surviving, you can rest assured these numbers are far more believable than any MTBF rates published by the manufacturers. This is real world testing as opposed to computed estimates, which backed by bean-counter calculations of replacing failed retail drives with bill-of-materials cost drives carrying only a residual warranty.
Personally, I would expect poorer durability in a home or office computer simply due to power cycles. BackBlaze probably never turns theirs off, even to replace a drive. But I'm sure as hell not going to buy from the high-failure-rate brands even if my use case would be different then theirs.
No, you are mistaken. I've always had this sig.
(Score: 3, Insightful) by Hairyfeet on Friday November 18 2016, @04:42PM
I wouldn't say the data isn't without use for those running server farms but I WOULD say its pretty useless for the average use case of a consumer hard drive. What they are doing is buying COTS consumer drives and then slamming them hard 24/7 in a role that they were never intended to do and publishing the results. This is great for those that want to build a NAS or server farm with lots of storage on the cheap but its not gonna tell us how a consumer drive used in the way it was designed to be used is gonna hold up.
Look at it this way....it would be like saying you can tell which small car is the best by taking a bunch of them straight off the lot and putting them in a 10,000 mile off road competition. While this would tell you which small cars happen to be good at 10,000 mile off road competitions that doesn't change the fact that you are taking something never intended to do the job you have given it and then trying to judge it based on tasks it was never designed to do.
And just look at the drive cages they use with multiple fans cooling each individual drive....when was the last time you saw that level of active cooling in your average HP or Dell tower? This kinda stuff really matters ya know, for example they like Seagate but any shop guy like me will tell you Seagates are good if and only if you give them a LOT of active cooling. If you stick it in a case with a 240mm fan blowing across it? Then even their most notorious drives like their 1.5TB and 3TB Barracuda drives will last for a very long time. OTOH if you put that very same drive in your average poorly ventilated Dell or HP mid-tower? You'll be lucky if it makes it out of warranty before it shits itself because ever since they bought Maxtor any shop guy will tell you Seagates just don't handle heat like a Samsung or Hitachi can.
So while this will tell you how these drives can do in this specific use case I really don't see it being useful for figuring out average MTBF for these drives used in a consumer box with the kinds of loads they were designed for.
ACs are never seen so don't bother. Always ready to show SJWs for the racists they are.
(Score: 5, Insightful) by frojack on Friday November 18 2016, @10:28PM
slamming them hard 24/7 in a role that they were never intended to do
I have no reason to think this is any different than my drives in my Home server of a typical small business server.
What do all of these have in common:
Such machines are never turned off. (This extends the life of drives - power up is way harder than constant spin).
They run 24/7, but each is probably worked about the same amount, because people sleep.
The vast majority of sectors are written once, or a very few times. Write once Read mostly.
The use case isn't all that different. And the "SLAMMING" is mostly in your imagination.
Also your main tenant, that hard use is no measure of durability for light use, has been disprove in just about every industry it been tested. Which, in the physical world is just about everything. Including cars.
No, you are mistaken. I've always had this sig.
(Score: 2) by Hairyfeet on Saturday November 19 2016, @01:04AM
Sorry but they are SLAMMING the drives as you don't buy assloads of storage if you aren't using assloads of storage! Know how much space is typically free on your average desktop PC? I can tell you, its 60%, that means nearly half of the drive has never been written to and the heads aren't having to traverse the whole drive equaling less wear on the motor, do you know how much free space is on these drives? Considering how many drives they are buying my guess is not much.
And again look at their drive cages, they have a lot of active cooling...how many fans do you see blowing on the HDD in your typical Dell or HP? Again I can answer that question, the answer is NONE, no fans are blowing on the HDD of your typical mid tower, they have a CPU fan and a rear exhaust fan and that is it.
So sure it'll help you if you are building servers or NAS, that was my point, just as taking those small cars and running them for a 10,000 off road rally will tell you which car to buy if you want to run a 10k off road rally, but that isn't gonna tell the person who just wants a small car to go back and forth to work each day which one to buy, the data just isn't relevant.
ACs are never seen so don't bother. Always ready to show SJWs for the racists they are.
(Score: 0) by Anonymous Coward on Saturday November 19 2016, @03:56AM
You friendly neighborhood word Nazi here.
"your main tenant" > This really confused me for a few seconds, and had me wondering what do people living at GP's house have to do with this, maybe they're using his servers?
You probably meant "tenet [dictionary.com]".
The word tenet , defined here, should not be hard to pronounce. For speakers of American English, say the number ten, then add the pronoun it , and you have tenet , pronounced (ten ʹ it). Unfortunately, there is a similar-looking and similar-sounding word in English that is much more common—the word tenant , meaning someone who rents and occupies an apartment, office, etc. This word is pronounced (ten ʹ ənt), and its pronunciation is frequently used in error by people who intend to say tenet . Because both words involve sequences of the same letters t and n —both of which are pronounced with the tongue in the same place, touching the upper palate—it is easy for the extra n of the more common word tenant to creep into the pronunciation of tenet . With care, one can learn to pronounce these two words differently and appropriately.
Back on-topic, there may be some to what GP says about different use-cases, but the results could swing either way. I'd also say that the very smooth and clearly delineated slopes of the graphs suggest that data is very representative :)
In my particular use-case, a desktop tower with several HDDs, ~95% full on average (whenever I run a new simulation, I have to find some old data that I don't need to analyze anymore so it can be discarded...), this data can be very helpful. Thanks to BackBlaze for publishing it and the submitter and editors for pushing it to SN! It's been a while since I last bought an HDD, so I forgot about this invaluable information :)
(Score: 2) by krishnoid on Friday November 18 2016, @07:48PM
Then again, I'm only a bear, and bears are said to have remarkably small brains.
I just had the strangest image go through my head.
"Gee Yogi, are you sure we can run this whole datacenter by ourselves?"
"Fret not, my diminutive friend! It's just a bunch of hard disks under vendor support. What could go wrong?"
(Score: 2) by bob_super on Saturday November 19 2016, @12:41AM
The main problem is that the backblaze stats do not take into account the hidden HDDs. Whether Backblaze is failing to reach them, or they're just lying on their lifespan, the end result is a clear skew in the drives' dying expectations.
In the end, you get surprised when you open the box, and you don't get the drive everyone had told you that you were getting.