Slashdot Mirror
8TB Drives Are Highly Reliable, Says Backblaze (yahoo.com)
An anonymous reader writes from a report via Yahoo News: Cloud backup and storage provider Backblaze has published its hard drive stats for Q2 2016. Yahoo News reports: "The report is based on data drives, not boot drives, that are deployed across the company's data centers in quantities of 45 or more. According to the report, the company saw an annualized failure rate of 19.81 percent with the Seagate ST4000DX000 4TB drive in a quantity of 197 units working 18,428 days. The next in line was the WD WD40EFRX 4TB drive in a quantity of 46 units working 4,186 days. This model had an annualized failure rate of 8.72 percent for that quarter. The company's report also notes that it finally introduced 8TB hard drives into its fold: first with a mere 45 8TB HGST units and then over 2,700 units from Seagate crammed into the company's Blackblaze Vaults, which include 20 Storage Pods containing 45 drives each. The company moved to 8TB drives to optimize storage density. According to a chart provided in the report, the 8TB drives are highly reliable. The HGST HDS5C8080ALE600 worked for 22,858 days and only saw two failures, generating an annualized failure rate of 3.20 percent. The Seagate ST8000DM002 worked for 44,000 days and only saw four failures, generating an annual failure rate of 3.30 percent." For comparison, Backblaze's reliability report for Q1 2016 can be found here.
UPDATE 8/2/16: Corrected Seagate Model "DT8000DM002" to "ST8000DM002."
UPDATE 8/2/16: Corrected Seagate Model "DT8000DM002" to "ST8000DM002."
...they use helium in the drives, so all your music sounds like Alvin and the Chipmunks.
I think it's more about HGST drives being highly reliable
Reliability is not so great an issue with raid systems being what they are today. What the bean counters fail to consider is the cost in man power required to replace seagate drives on a constant basis. Not just in the racks but process RMA's or the proper destruction and disposal of drives which may contain sensitive data.
I wonder how those numbers would look if other vendors were offered an equal analysis period. I know WD was mentioned but it didn't appear they got equal share.
Also: First. :)
~ People that think they are better than anyone else for any reason are the cause of all the strife in the world.
If your 8 TB hard drive is highly reliable, can you rely on your network connection and operating system to spy on everything you do?
Which city is Ed Snowden in presently?
Totally not trying to be pedantic, but the Seagate model they reference should actually be the "ST8000DM002"
These are all platter drives, but you can only discover that in the comments at TFA.
There are so few 8GB HGST drives, and they're so new, that the current data about them is statistically insignificant/unreliable, as is any model with less than 500 units and 200k drive days.
Just buy a price point and run in your production gear? Never take a peak at the firmware, or drive performance over time? Errors handled correctly? Wow dude. That takes guts.
I guess if you're cloud, you don't give a crap about corruption, but if you're looking at over 5% AFR's, you might want to take a closer look. You've probably got a firmware problem, and they probably have a fix for it. It may be something specific to your work load. I'm sure you buy enough drives from them so you can at least get your distributor to set something up. Seagate and HGST have some really good engineers that would probably want to know about the fallout before a year has passed. That RMA thing can get swallowed in the numbers.
"...the company saw an annualized failure rate of 19.81 percent with the Seagate ST4000DX000 4TB drive"
A failure rate of almost 20% in a data center? Geez, that's pathetic.
A temperature-controlled environment, clean power, low shock and vibration, and 1 out of 5 still fails? Remind me never to buy Seagate. Oh, wait, I already vowed never to buy another Seagate- about 10 years ago after experiencing their unequaled propensity to die fast and hard.
Maybe other people have had better luck with Seagate than I have, but for me they've always been disappointing.
Just cruising through this digital world at 33 1/3 rpm...
It seems the ST8000DM002 is a desktop drive. I've had 3 of the Archive variants (ST8000AS002), and they all failed within a week of use. The third time I got a refund.
If you've got 3,000 drives at home to come up with directly home applicable numbers, then please share them.
This is mostly useful to compare models vs models as the environment is kept the same.
It's completely legitimate to say model X is more reliable than model Y, it's not valid to say model X has a Z% failure rate in a home environment however.
I presume there's some detail I'm missing here since we did not have 8 TB hard drives 120 years ago.
In a bit of shameless internet panhandling, I accept Litecoin Donations at Lbd2oH9QsthD1GfuUXPyka12YxvWJYnBVf
I've got a cool story, bro. We put a NAS device online in our new data center with 4 Seagate Archive 8000 drives in it, and 2 of them died within 24 hours, trashing the RAID array. Thankfully, since it was a new NAS, it wasn't a big deal.
"Our two-party system is like a bowl of shit looking at itself in a mirror." - Lewis Black
Who measures uptime in drive days? That is like saying man hours or multiplying uptime by the number of disks in an array. Where does it end?
Maybe I'm old but damnit I want min / max / avg - in DAYS, per drive. Can we please stop changing units on everything?
Unless we're talking years of uptime, these numbers don't come close to my desktop hard drives let alone something MADE FOR A DATA CENTER.
Anyone who uses a Helium filled Hard drive won't want to be reminded what happens when the Helium leaks out ...
If you've got 3,000 drives at home to come up with directly home applicable numbers, then please share them.
This is mostly useful to compare models vs models as the environment is kept the same.
It's completely legitimate to say model X is more reliable than model Y, it's not valid to say model X has a Z% failure rate in a home environment however.
I would agree with you. The environment is the same for all devices. The sample sizes vary, but I think the useful numbers come from the very large sample sizes.
It's interesting to note that Backblaze tracks failures across a specific model, but they do not provide numbers on failures within a specific model. For example, did a large number of devices fail and used the same firmware? or fell within the same manufacturing date range? or came from the same country or specific serial number range? Another way of saying that is: Can Backblaze do a better job of isolating their failures down to a more specifc identifier??? Some of the needed info is actually printed on the drive labels and not accessible via SMART, like manufacturing date & country of origin.
I noticed that Backblaze uses some WDC RED drives. I cringed when I read that. I believe that WDC designs RED drives from 1 to 4TB for systems with 5 drives or less. I am guessing here on that design limit. It must be "common sense" to expect failures from those drives when used in high density platforms like the pods at Backblaze. I mean using a device in an environment for which it was not designed by the manufacturer is like expecting a Ferrari 308 to be a great "off road" race car for the Baja 1000!!!
As an owner of 30+ RED drives in that size range, and all installed in homemade NAS systems of 5 drives or less, (knocks on wood...many times) my own experience has been extremely positive: no failures across 10,000 hours per drive. I have at least 300,000 hours of drive "uptime" without a failure for WDC RED 3 & 4 TB drives, but I do have spares stored..."just in case".
Here's what I find is important for getting the longest life out of a hard drive: (1) extremely secure and stable hard drive mountings; (2) all of my systems are on "prosumer" or "business" branded UPS systems (Eaton, in my opinion, is one of the best UPS out there); (3) OS--NAS integration (Linux is a great OS) such that NAS goes into automatic "safe" shutdown when the UPS runs on batteries for more than 5 minutes; (4) use drives designed for NAS or RAID usage (likely to have appropriate TLER tweaks in firmware); (5) check all drive firmware and burn-in all drives before placing them into "service" (avoids many premature failures and problems due to shipping).
I wonder if some of Backblaze's issues with WDC 4 TB and below drives was a lack of firmware upgrades. I distinctly remember there being a firmware upgrade that was not needed by my 4TB REDs but a number of my older 3TB REDs needed that upgrade. I forgot what WDC changed in that firmware upgrade, but I remember thinking it was necessary to do the upgrade work based on reading the "sparse" release notes from WDC that I could find.
Come back in 3 or 5 years and tell me out of all the 8TB sold in 2016/2017 just how many are still functional and THEN what the failure rate is/was.
My "prediction" is it will most likely be that there is an 70% failure rate with Seagate being the top offender.
I believe that WDC designs RED drives from 1 to 4TB for systems with 5 drives or less.
WD RED drives are available in 6TB and 8TB. Regular drives are 5400RPMs with 64MB cache. Pro drives are 7200RPMS with 128GB cache.
The 8TB Archive drives with shingled recording are not suitable for RAID arrays.
If it's working for them in their packed in boxes with crap airflow and really poor heat transfer then it will work even better in conventional file servers with hot swap drives at the front and a heap of airflow.
Take it with a grain of salt when Backblaze say a drive is crap since it may only be crap in their very hostile environment, but if they didn't break it then it's very likely to work well anywhere.
Is that a typo or is there really 2000 times the cache?
You wot mate?
Seagate Archive drives are designed for cold storage, as they say 6 times on their web page for the drive. If you don't know what "cold storage" means, it means "not RAID".
So, you build a RAID array out of drives designed for "not RAID", and they started failing on you. And this is somehow Seagate's fault? The mind boggles.
Socialism: a lie told by totalitarians and believed by fools.
Yes, a typo. But those pro drives are really expensive.
Drives of that size are no longer limited to gimped "archival" roles.
On the one hand, a drive is probably likely to be more reliable when you pamper it and don't really do much with it. On the other hand, I've had plenty of Seagates fail in just that kind of use case.
Gimped archive disks? Who cares if they are reliable or not?
A Pirate and a Puritan look the same on a balance sheet.
> The HGST HDS5C8080ALE600 worked for 22,858 days and only saw two failures
The Maths on this summary are worded really strangely. I am guessing they mean N drives for X days = 22,858 days, but it sounds like their 8TB drives were powered on 62 years ago and ran great except for two that died from the Germans bombing their data center in WW II. Oh wait, that was 72 years ago, so it must have been the 4 TB drives that died. My mistake. Really though, 22,858 days needs to be phrased like "man-hours", "man-years", etc. How about 22,858 unit-days?
I had a 1st Gen Seagate 80GB SATA fail last month after 11 years and change, of 24/7 daily operation and very few power-off cycles.
Political debates have me rolling my eyes so much I think I got optical whiplash. I should sue. - Foamy The Squirrel
are directly attributable to the controller boards. I have a 3TB from one of their USB drives that failed on me (It would just randomly start writing garbage to the drive, low level format would fix it, but reading back after would return garbage.) Another is a 500GB still in service that will start throwing read errors when the temp sensor reads around 45-50C (common during the summer here if the cooling is impeded the drive will heat up and begin malfunctioning, requiring both a cool off period and a full power-cycle to come back up properly. Oftentimes it results in data corruption across much of the directory tree resulting in most of the file system ending up in the lost+found folder, and a reinstall being needed to fix the system.
That particular system has a mix of third party drives in it, none of which see failures except for the seagates. (A diff seagate drive with ext2 filesystems does not see similiar corruption, although it is also accessed less frequently.)
I'm an independent white-box NAS guy, and with the exception of the truly awful 1.5TB Seagate drives from 2008-2009 or so, I have not had any significant problems with them. I've got a few thousand 3 to 8 TB drives deployed with my clients, most of them cheap consumer drives (not even the "NAS" editions), and the annual failure rate is roughly 2% across all brands. This has been consistent for many years and I factor these stats into my costs and warranty projections. I have
The thing that bothers me about Backblaze, and the reason why I have a very hard time taking their results seriously, is the way they design their pods. They take a custom fabbed chassis, then fill it with the most ghetto components known to man: SATA port multipliers, ultra-low-end HBAs, dual "gamer" power supplies, very substandard cooling, and until recently they used super sketchy desktop boards. It's only last year that they finally changed the board for a Supermicro, primarily to get 10GbE very cheaply. For that same money, you can buy a ready-made 60-bay Supermicro chassis with redundant power and SAS - and a warranty. Hell, I bet SM would deliver directly to Backblaze's doorstep *and* give them a friendly discount.
Anyway... epic digression aside, when people ask me which brand is better, I tell them to buy whichever has the best warranty. A hard drive *will* die, the question is when, so the only logical course of action is to plan around its inevitable demise by keeping backups and redundancies, and learning the ins and outs of the RMA process.
-Billco, Fnarg.com
I end up reading the EXACT same comments and arguments.
What I've learned from reading the comments here is that people are just as clueless when it comes to storage reliability as they ever were, and are just as capable of throwing the baby out with the bathwater as at any other time.
Dear Slashdot: Never change.
Kid-proof tablet..
If ever you thought that nerds were more scientifically minded than others, just ask them about their hard-drive preferences and watch them wax anecdotal.
By far the most used drive in the above report is the 4GB Seagate ST4000DM000: ~34,000 drives with a 2.7% failure rate. Two Toshiba's and one WDC show failure rates of nearly 9%. HGST is the only manufacturer with consistently sub-5% failure rates.
Aren't we all building raid arrays with non-raid disks for SMB and home use nowadays anyway ?
If you're being pedantic and take RAID to mean "Redundant Array of Inexpensive Disks" then the Seagate Archive is in fact the most ideal candidate.
I wouldn't though but that doens't mean he shouldn't.
No. The archive disks aren't designed for a long duty cycle. They are meant to have data dumped onto them and then work as a read only disk.
The constant usage of a RAID array will cause drive failure via thermally induced URE in short order
Chas - The one, the only.
THANK GOD!!!
The enterprise drives always seem heavier to me.
Drives designed for RAID use typically have different firmware which react differently to issues - RAID friendly drives react quicker to failures, meaning they are less likely to fail the RAID over correctable errors. Put a drive not intended for RAID use in an array and you will see more failures over drive level correctable errors.
Archival disks are one of those drives you will see this issue with.
the most unreliable.
That is why you buy in the sweet spot for best value and let someone else prove new technologies and HD densities for you..
My local recycler has two machines that shred drives like a paper shredder. It doesn't matter what type of drive. You pour drives in the top and flakes of metal fall out the bottom within 3 seconds. Not even SSD chips survive these machines. Encryption, overwriting, drilling, are all pointless wastes of time. Shred em and forget about it.
When I perform a new system roll out and old systems are being recycled, I pop out the drives and the recycler shreds them en mass right in front of me. 50 drives takes ~5 minutes.
The machines are ridiculously simple so when the recycler told me that they cost $15,000 to $20,000 each I didn't believe him. But I later verified it to be true. Even the cheap little portable "wood splitter" type device that presses and splits two or three drives at a time is over $4,000
SMR drives are different - the S is for Shingled. It's an oddball recording technology that requires an entire track be written to change any block. They're really the worst choice for random write patterns.
Socialism: a lie told by totalitarians and believed by fools.
and the Japanese brands are highest quality, it's just like their previous reports. HGST, while being owned by an American company, is still doing all their research and development in Japan, by Japanese engineers, and this is likely the key difference that makes the HGST drives come out so far above the American drives.
Uhm... No. SMR drives only have very huge sector size, as far as the filesystem is concerned. If anything, the performance would be incredibly bad, but no reason to fail. Not within 24 hrs.
Why would anybody buy a spinning hard drive today so slow and solid-state drives are so cheap . I have a terabyte SSD drive in the new MacBook has read and write speeds close to a gigabyte a sec I have an eight-year-old MacBook, that before I swapped the hard drives for ssd had a 22 MB a second read write speed. Samsung has dominated the market micron might have a chance but anybody making spinning hard drive is wasting time.it's time the world implements a lifecycle tax. Meaning if the product last hundred years there's no tax if it lasts one day its thousand percent we don't need anymore happy meal toys
These are the same guys that claim HGST hard drives are reliable. Yeah, no thanks!
OpenZFS has been working to become aware of shingled storage. The CoW nature of ZFS already plays well with shingled recording, but it will become much better once the FS is aware of the layouts. In theory it's not much work, in practice, it's a lot of refactoring.
To change any earlier block. Changing earlier data requires later data to be re-written because the write head is wider than the read head. As long as you append data, you're fine. There in lies the rub. How do you know if you're near the front or back of a shingled region? If it's always per track, then that information is available. Even then, most/all file systems don't care. OpenZFS will care in the future. CoW nature plays well with being able to almost always append to these regions, reducing the amount of re-writing.
If you don't know what "cold storage" means, it means "not RAID".
Well, it would be fantastic if they would mention that in one place or another. Instead I get lines like this:
Enjoy peace of mind with a drive engineered for 24×7 workloads of 180TB per year
Store your data faster with a SATA 6Gb/s interface that optimizes burst performance
Have confidence with a drive that provides reliable, low-power data retrieval based on Shingled Magnetic Recording (SMR) technology
Yes, this sounds like exactly what I'm looking for when I store my backups. I don't need to write to them often, really only once a day. Reliable retrieval sure would be nice though.
learn how these affordable, high density drives can meet your needs for long-term, cold storage that's quickly and readily available online.
That's great, I would like my long-term backups quickly and readily available online when needed.
Seagate Archive HDD has won the "Product Award of 2015" in the 3.5" segment by Kakaku.com
Ah, sounds like they thought it was the best 3.5" drive.
...one of the best all around hard drives on the market.
That's good, I'd like one of the best all around hard drives on the market.
Seagate Archive HDD 8TB: A lot of TBs for a relatively small investment.
Yes, I need a lot of TBs.
The Seagate Archive HDD 8TB is a high capacity, energy efficient, and lower cost hard drive for active archive purposes.
Active archive sounds like backup storage. This must be for me.
The drives are intended for use in large-scale data centers where density, power consumption, data integrity and data retrieval are paramount.
That's good, because I'm going to put these in a large-scale data center where data integrity and data retrieval are paramount.
Best fit applications:
Online archiving
Large data object storage
Big data cold storage
Cloud active archive
Web-scale archiving
All of those buzzwords sure sound similar to "where you put your backups".
Delivering absolute lowest cost/TB along with the performance and reliability required for massive scale applications, the new 8TB HDD is ideal for meeting the needs of our enterprise and service provider customers who demand optimized hardware and the cost structure needed for massive scale out.
Yes, massive scale, like you would find in a redundant array of disks.
But, if I pull up the data sheet, then it includes this footnote which is missing from the same section on the web page:
Archive HDDs are not intended for surveillance or NAS applications, and you may experience lower performance in these environments.
By "may experience lower performance", I'm guessing they mean that if I put these in a RAID array and point my servers there as a backup location, then I can expect a 50% failure rate in 24 hours.
"Our two-party system is like a bowl of shit looking at itself in a mirror." - Lewis Black
"RAID friendly drives react quicker to failures, meaning they are less likely to fail the RAID over correctable error"
Your're referring to TLER - which used to be a tunable value until Seagate/WD started using it to differentiate enterprise/domestic drives (it dictates how hard a drive will try to recover sector errors before marking them bad and moving on)
On the other hand to your example, if you put a RAID-friendly drive in standalone use and there's a sector issue you're far more likely to lose data.
It would be interesting to know if the TLER is tuneable on these drives (it isn't on lower capacity STx000DM-x drives), but given a 200%+ failure rate in the warranty period on Seagates's DM001 drives (2 and 3TB) I would still be very wary.
24 hours? You're having a laugh.
No. No I'm not. Those drives simply don't have the features to survive in an array environment.
So, like an ordinary desktop drive (which is also missing those features), they'll eventually desync and fall out of the array.
If they tried to put the drives under load (like migrating the contents of one NAS to another), it's ENTIRELY possible that the drives died due to thermal excess (which is what happens when you run them for long periods of time).
And if they're packed in a small NAS box (think Synology DS1515, Drobo, etc), all up tight to one another? They'll cook themselves in short order.
Again, SMR Archive drives ARE NOT meant to be run in RAID/NAS environments! PERIOD! Talk to the manufacturers. They'll tell you the same thing.
Chas - The one, the only.
THANK GOD!!!
Read what I have written it is very simple. Semiconductors behave differently when hot and that sometimes leads to failure. There is a lot of heat input from the mechanical side of the drives. If it can't be transferred away you get hot electronics no matter what you do on the electronic side.
Does that make sense yet?
Even if the drive was not intended to be used for RAID, or designed to spend most of its time sitting on a shelf, I would still expect the drive to last more than 24 hours. Heck, even writing 8TB of data to the drive once then reading it back would take a good fraction of that 24 hour "lifetime". Now, if the drive died after, say, 2 months then maybe your comment would apply.