The Reality of SSD Capacity: No-One Wants Over 16TB Per Drive
One of the expanding elements of the storage business is that the capacity per drive has been ever increasing. Spinning hard-disk drives are approaching 20 TB soon, while solid state storage can vary from 4TB to 16TB or even more, if you're willing to entertain an exotic implementation. Today at the Data Centre World conference in London, I was quite surprised to hear that due to managed risk, we're unlikely to see much demand for drives over 16TB.
Speaking with a few individuals at the show about expanding capacities, storage customers that need high density are starting to discuss maximum drive size requirements based on their implementation needs. One message starting to come through is that storage deployments are looking at managing risk with drive size – sure, a large capacity drive allows for high-density, but in a drive failure of a large drive means a lot of data is going to be lost.
[...] Ultimately the size of the drive and the failure rate leads to element of risks and downtime, and aside from engineering more reliant drives, the other variable for risk management is drive size. 16TB, based on the conversations I've had today, seems to be that inflection point; no-one wants to lose 16TB of data in one go, regardless of how often it is accessed, or how well a storage array has additional failover metrics.
Related: Toshiba Envisions a 100 TB QLC SSD in the "Near Future"
Samsung Announces a 128 TB SSD With QLC NAND
Related Stories
Forget the 60 TB SSD. Toshiba is teasing a possible 100 TB SSD:
The Flash Memory Summit saw Toshiba deliver a presentation about quad level cell (QLC) technology – adding substantially to the prospect of a product being delivered in the "near future". We have heard about this QLC (4bits/cell NAND technology) quite recently.
After Seagate tantalised us with a 60TB SSD, along comes Toshiba with a 100TB QLC SSD concept.
Flash Memory Summit attendees saw Toshiba presenters put flesh on the bones and envisage a QLC 3D SSD with a PCIe gen 3 interface and more than 100TB of capacity. It would have 3GB/sec sequential read bandwidth and 1GB/sec sequential write bandwidth. It would do random reading and writing at 50,000 and 14,000 IOPS respectively. The active state power consumption would be 9 watts, the same as a 3.5-inch, 8TB SATA 6Gbit/s disk drive, while the idle power consumption be less than 100 mWatts, compared to the disk drive's 8 watts.
Even if the "near future" isn't so near, or the final capacity does not end up at around 100 TB, it is still interesting to see 3D NAND technology enabling a serious push for 4-bits-per-cell NAND, which would normally face endurance issues.
Samsung will use QLC NAND to create a 128 TB SSD:
For now, let's talk about the goods we'll see over the next year. The biggest news to come out of the new Samsung campus is QLC flash. Samsung's customers set performance and endurance specifications and don't care about the underlying technology as long as those needs are met. Samsung says it can achieve its targets with its first generation QLC (4-bits per cell) V-NAND technology.
The first product pre-announcement (it doesn't have a product number yet) is a 128TB SAS SSD using QLC technology with a 1TB die size. The company plans to go beyond 16 die per package using chip stacking technology that will yield 32 die per package, a flash industry record.
NAND revenue has increased 55% in one year.
Previously: Seagate Demonstrates a 60 TB 3.5" SSD
Toshiba Envisions a 100 TB QLC SSD in the "Near Future"
Western Digital Announces 96-Layer 3D NAND, Including Both TLC and QLC
Toshiba's 3D QLC NAND Could Reach 1000 P/E Cycles
(Score: 2, Interesting) by RandomFactor on Friday March 15 2019, @12:03AM (1 child)
They Call Me Nobody [youtube.com]
В «Правде» нет известий, в «Известиях» нет правды
(Score: 2) by bart9h on Wednesday March 20 2019, @03:03AM
Nobody was faster than him!
(Score: 4, Insightful) by Anonymous Coward on Friday March 15 2019, @12:12AM (10 children)
If you have 16tb on a drive, and you back it up, and the drive dies, then you don't lose any data. Same goes for a 160tb drive. Do not underestimate the ability of SoylentNews users to collect porn.
(Score: 0) by Anonymous Coward on Friday March 15 2019, @01:50AM (1 child)
Another dog-slow and short-lived QLC idiot toy?
If a "storage device" needs constantly be backed up onto another reliable one, let's call the second one "storage" and the first "cache", and stop reinventing the square wheel.
(Score: 2) by Immerman on Friday March 15 2019, @02:26PM
>If a "storage device" needs constantly be backed up onto another reliable one,
Except that's true of *every* storage device, regardless of reliability, because there's no such thing as a truly reliable device.
There's only two kinds of data in the world: that which you've already backed up to an independent storage device (preferably off-site), and that which you are obviously willing to lose.
(Score: 2) by driverless on Friday March 15 2019, @03:38AM (2 children)
If I had a 16TB drive and it died, I'd just re-download all the porn... uhh, restore all the data that was on it from public sources. So no big loss really.
(Score: 0) by Anonymous Coward on Friday March 15 2019, @05:25AM (1 child)
How fast can you download 16TB?
(Score: 0) by Anonymous Coward on Friday March 15 2019, @05:37AM
Torrents at 10 megabytes per second total? Filled in less than a month.
(Score: 0) by Anonymous Coward on Friday March 15 2019, @05:23AM (4 children)
Given the enormous amount of free porn on the 'net, why would one want to collect it? Just stream it when it's desired/required(?).
(Score: 0) by Anonymous Coward on Friday March 15 2019, @05:27AM (3 children)
If you want something, you should download a copy if you ever want to see it again.
Try streaming the Brenton Tarrant Facebook live video. Possible, but not so easy.
(Score: 0) by Anonymous Coward on Friday March 15 2019, @12:02PM
Okay. If that's the kind of stuff you masturbate with, I guess I can understand.
There's *plenty* of (pretty kinky) stuff I'm into online, so I'm set thanks. I even get to have sex with actual women on a regular basis. The same one even. I'm sorry. Does that make you jealous?
Shooting people (at least with guns) doesn't turn me on at all.
(Score: 3, Touché) by DannyB on Friday March 15 2019, @02:59PM (1 child)
True. I am concerned [wikipedia.org] that I might not be able to find Rick Astley song "Never Gonna Give You Up".
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 2) by NotSanguine on Friday March 15 2019, @03:50PM
Yep. I'm really worried about that too [bit.ly]. Hopefully, we can make sure that doesn't happen.
No, no, you're not thinking; you're just being logical. --Niels Bohr
(Score: 4, Funny) by Subsentient on Friday March 15 2019, @12:12AM (3 children)
I get to race such drives and see how long they take to complete a "cat /dev/zero > /dev/sda"
"It is no measure of health to be well adjusted to a profoundly sick society." -Jiddu Krishnamurti
(Score: 5, Funny) by The Mighty Buzzard on Friday March 15 2019, @12:18AM (2 children)
You know, I used to do that. Then I found out how much faster /dev/null's write speed is and never looked back.
My rights don't end where your fear begins.
(Score: 1) by jbruchon on Friday March 15 2019, @02:21AM (1 child)
You're welcome. [youtube.com] Relevant bit is at 2:00-2:15, but watch it all for maximum enjoyment.
I'm just here to listen to the latest song about butts.
(Score: 2) by DannyB on Friday March 15 2019, @03:03PM
That ignorant video suggests that you cannot shard /dev/null in order to gain superior write performance.
Here are helpful tips for the skeptics.
Make /dev/null scalable in case you are doing too much IO to /dev/null!
Shard /dev/null into multiple nodes.
Create a /dev/null1 like this:
$ sudo mknod -m666 /dev/null1 c 1 3
$ ls -la /dev/null*
Or a loop to symlink:
$ sudo for i in count(2-19) ln -s /dev/null$i /dev/null
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 5, Interesting) by DrkShadow on Friday March 15 2019, @12:13AM (5 children)
You could say the same thing about 1GB of data 20 years ago. It's all about price and performance.
20 years ago, how long would 1GB have taken to transfer? a while.. not so bad. Right now, how long does 1TB take to transfer? a while.. not so bad. 20 years from now, how long will 1PB take to transfer? A while.. probably not so bad. We already have NVMe. It's linking in with the system bus. Throw RAID controllers in that don't even have to access the system bus, 100Gb+ disk links, and 1PB will take a while, but it won't be so bad. (UDMA33 -> UDMA133 -> SATA -> SATA2 -> 12Gbps SATA3, so -> 24Gb -> 48Gb -> 96Gb -> 200Gb -> 480Gb? 1PB at 480Gb takes.. about what a 10TB hard drive takes now, if I did my approximations right.)
In 20 years, no one will want 4PB disks. They're just too big for the performance vs risk vs cost tradeoff.
16TB, now? It's a lot of data. It can be used for local caching, where if the data goes it doesn't really matter. Why are you caching such large volumes with SSD's, when you can't quickly access _all_ of it? hmm. Performance sucks for that capacity, unless you're talking NVMe, which doesn't have that capacity. Sata3 is too slow for it.
(Score: 0) by Anonymous Coward on Friday March 15 2019, @12:38AM (2 children)
I didn't notice a difference in real life read/write times between a usb 3.0 external SSD and a NVMe m.2 SSD. Did I do something wrong?
(Score: 3, Informative) by DrkShadow on Friday March 15 2019, @02:54AM (1 child)
Yes (https://www.newegg.com/Product/Product.aspx?Item=N82E16820250099 [newegg.com]).
The USB 3.0 bus is limited to 5Gbps. Most USB SSD's are sata, adapted to USB, so at most SATA3 (6Gbps) for USB-C 3.2 Gen 2 (ugh, that article). The disk posted above, PCIe 3.0 x4, is rated at 3400MB/s (27Gbps). I have little doubt that it will actually give you this performance.
You get what you pay for. It costs 2.5x as much as other NVMe disks of the same capacity.
(Score: 0) by Anonymous Coward on Friday March 15 2019, @03:10AM
Well I tried it. I put my data I wanted to read on the NVMe m.2 drive and timed it. Then I did the same from the external, it made almost no difference. Maybe for my task the actual reading of the data wasn't the bottleneck or something.
(Score: 3, Interesting) by jbruchon on Friday March 15 2019, @02:28AM (1 child)
The problem with really huge disks is throughput. I'd gladly have a 4PB disk, but the amount of the disk that can be moved around per minute is a major problem. Sure, incremental backup can skip most of the problem temporarily, but an initial backup or restore of 10TB of data at 100MB/sec average throughput takes more than a full day to pull off. If you were to chart the size of drives divided by the maximum sequential throughput, the problem would stick out like a sore thumb. 160GB desktop drives from 10 years ago can push something like 40-50MB/sec while a brand new 3TB 7200RPM drive maxes out at around 180MB/sec, but that's 3-4x the throughput at 9-10x the size. The ever-increasing layering of flash memory in SSDs and the slowdowns that come with it only make this problem worse; if a 10TB SSD has slower sequential throughput than a shucked 10TB WD EasyStore and most of my data is decently large files, I'm not going to go with the SSD, I'm going to shuck those bitches till I'm swimming in plastic carcasses.
I'm just here to listen to the latest song about butts.
(Score: 1) by shrewdsheep on Friday March 15 2019, @03:25PM
I am not sure your argument is relevant for the data center. Whether a single 16Tb drive fails or 10 1.6Tb drives does not matter for the data center. There will be an average file-restore capacity needed in the network and it does not matter to which drives it goes. Smaller data centers could see some discretization effects as could some special applications. Disk size changes have to go hand in hand with changes in network infrastructure and I speculate that we are rather running into a disconnect at the moment. Networks will adjust and disk size will happily grow again.
(Score: 4, Insightful) by The Mighty Buzzard on Friday March 15 2019, @12:25AM (17 children)
If you're relying on a non-backed-up drive to never fail in a big commercial environment, you have no business making tech decisions in the first place. If you're not, you aren't going to lose any data if it fails and you're going to save tons on power bills by going with the (almost) largest drives you can find.
My rights don't end where your fear begins.
(Score: 4, Insightful) by NateMich on Friday March 15 2019, @12:35AM (9 children)
You may not "lose" the data, but the real issue they are probably talking about is that any given machine is out of service until the data is put back.
Of course, a better system is one where these drives are in an array, and when one fails you just change it out and keep going. No downtime required.
(Score: 2) by PartTimeZombie on Friday March 15 2019, @12:41AM (5 children)
Is there any commercial application where a drive would not be part of an array? Maybe as an offline backup I suppose.
(Score: 3, Insightful) by The Mighty Buzzard on Friday March 15 2019, @01:08AM (4 children)
Small businesses. Anyone who moves enough cash to afford someone who knows how to set up an office IT structure has no excuse for not doing so though.
My rights don't end where your fear begins.
(Score: 2) by PartTimeZombie on Friday March 15 2019, @01:35AM (3 children)
Now you point that out I've realised you're right of course.
In fact I have turned down a job setting up a server for a small business because it very quickly became clear that they didn't want to spend enough to do it properly.
I probably wouldn't have been paid enough to make it worth my while either.
(Score: 2) by The Mighty Buzzard on Friday March 15 2019, @03:52AM (2 children)
Depends on how important the server was to their business. Or at least how important they thought it was. You can occasionally get a small business that has a genuine need for a server and knows it but either quality is somewhat rare. Most can get by just fine with like half a dozen low-end-workstation-spec boxes or less but the ones that can't will find that out sooner or later.
My rights don't end where your fear begins.
(Score: 3, Insightful) by PartTimeZombie on Monday March 18 2019, @12:35AM (1 child)
Like every job I've heard of, the server would be incredibly important, only not worth spending any money on. ;-)
(Score: 2) by The Mighty Buzzard on Monday March 18 2019, @03:17AM
And that's the damned truth.
My rights don't end where your fear begins.
(Score: 3, Insightful) by The Mighty Buzzard on Friday March 15 2019, @01:12AM (1 child)
Like I said, if it's important data that's only on one drive, you failed right from the start. RAID and two sets of backups should be the bare minimum for most anything.
My rights don't end where your fear begins.
(Score: 3, Interesting) by AthanasiusKircher on Friday March 15 2019, @02:03AM
Yes, and if you actually care about data integrity in RAID and in those backups, you probably want ECC RAM and a filesystem that can detect and correct random errors and bitrot, like ZFS for example.
Once we're talking about data the size of 16 TB, data degradation is likely to happen over time statistically, even due to random bit flips.
(Score: 3, Informative) by darkfeline on Friday March 15 2019, @03:48AM
If you're talking about personal machines, restoring 500 GB to a 16 TB drive isn't any slower than restoring 500 GB to a 1 TB drive. If the 16 TB isn't too expensive it would definitely be a good option for data hoarders.
If you're talking about servers, it's all in the cloud and datacenters. No one cares if a drive goes down. Hell, regular disaster testing involves the simulated equivalent of knocking out random rows of racks of servers or entire datacenters and watching the entire system gracefully work around that. If you're one of those wondering what all this newfangled containers and Kubernetes and cloud fad is all about, THAT is what it's all about. Losing a 1 TB drive here, losing a few PB datacenter there, no big deal, the system shifts the containers and storage around, someone's going to be working harder to restore the lost redundancy, but otherwise business as usual.
Join the SDF Public Access UNIX System today!
(Score: 1, Insightful) by Anonymous Coward on Friday March 15 2019, @12:49AM (4 children)
This is all assuming that rebuild failures aren't a thing, and that the odds of them don't increase astronomically as the capacity of your drives increases.
(Score: 2) by The Mighty Buzzard on Friday March 15 2019, @01:10AM (3 children)
There are viable redundancies to eliminate most any worries available for every level of complexity you can name. Not using them is bloody stupid.
My rights don't end where your fear begins.
(Score: 1, Insightful) by Anonymous Coward on Friday March 15 2019, @01:17AM (2 children)
Unless of course you've already crunched the numbers and figured out running 2x8TB drives provides less risk of drive failure and subsequent downtime for your particular application. Redundancy is only as good as the time it takes to get your data back online, I can think of plenty of businesses where lower risk of drive failure and reliability is more important than drive density. I find it alarming that you think such a scenario doesn't exist.
(Score: 2) by The Mighty Buzzard on Friday March 15 2019, @04:01AM (1 child)
There is no properly designed situation where losing a drive of any size is a significant annoyance. If you find yourself in such a situation, you built your shit wrong. Now most people are not going to want to buy the largest drives because they're pricey as all fuck, they want something about one or two steps down for the sweet spot on price/dollar. That is for people who need to worry about up front costs though, not those who have the luxury of looking at lifetime costs. And, let's face it, if you're worried about up front costs, you're going to put your shit on someone else's hardware anyway.
My rights don't end where your fear begins.
(Score: 1, Insightful) by Anonymous Coward on Friday March 15 2019, @02:12PM
You are so wrong that it hurts and I really don't understand why you keep doubling down on such a stupid position. Even with perfect redundancy, losing a drive means decreased read/write speeds. There are several, very valid reasons why people who actually host large amounts of data would choose lower capacity drives; they are outlined in the article if you bothered to read it.
(Score: 3, Interesting) by Farkus888 on Friday March 15 2019, @02:57AM
The problem is raid rebuild times. There is a relationship between bus speed and size. SATA II had this problem at the end. Drives so slow but big you'd get additional failures and lose the array before a rebuild could complete. After all once one drive from a lot fails the others are likely close behind. Without doing the math I believe 16tb would be over that line on SATA III. As for NVMe, I doubt the size limit is that low. I particularly doubt it in an array type that supports 2 or more disks out safely. To be clear I follow tradition and did not RTFA.
(Score: 3, Funny) by DannyB on Friday March 15 2019, @03:46PM
Some people (not me thankfully) have PHBs to make tech decisions for them.
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 0) by Anonymous Coward on Friday March 15 2019, @12:31AM (2 children)
But it doesn't mean that at all?
>inb4 the potential harm is greater so the acceptable risk is lower
Just add another backup. Literally it's that simple, you can drive the chance of data loss down exponentially if you can make your backups failing independent events.
(Score: 0) by Anonymous Coward on Friday March 15 2019, @12:41AM (1 child)
Some people prefer predictable failures rather than statistically random independent ones.
(Score: 0) by Anonymous Coward on Friday March 15 2019, @01:23AM
The alternative to making the failures of your backups as independent as possible isn't making the failures predictable, it's making sure your backups all fail at the same time. You trade off minimizing the risk of data loss in favour of maximizing the chance that either all your backups work or none of them do. Moving towards non-independence is just approximating having a single backup, which you can do for cheaper by just having a single backup without any increase in risk of data loss, because it literally couldn't get any worse than perfect correlation.
(Score: 3, Insightful) by looorg on Friday March 15 2019, @01:04AM (10 children)
Sure I want one. But I assume "nobody" wants it due to the price. I have not seen any 16TB drives for sale yet but last I looked the 12TB and 14TB drives were about $700 each or thereabout. Totally not worth it for any or many home systems.
(Score: 2) by c0lo on Friday March 15 2019, @01:27AM
If they pay me enough, I'd be willing to save them the cost of warehousing one.
https://www.youtube.com/watch?v=aoFiw2jMy-0 https://soylentnews.org/~MichaelDavidCrawford
(Score: 2) by takyon on Friday March 15 2019, @01:47AM (8 children)
I think we've reached the point ($100) where we would rather have a 1 TB SSD than any HDD in a laptop. Of course, it helps that my primary laptop has a 500 GB HDD.
Desktops should have 1+ TB SSD and big HDD only as secondary and/or external storage. Maybe you can fill up a 1 TB SSD with applications alone, but it would take some doing.
Stuff like 96-layer 3D QLC NAND is going to lower SSD prices. Hopefully the downsides of QLC NAND aren't too apparent (cache, and frequent use to prevent possible data loss). And I'll be really amused if OLC (8 bits per cell) NAND is realized.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2) by c0lo on Friday March 15 2019, @01:53AM (4 children)
How filling up 1 TB SSD with apps helps me compiling faster that monster of the solution? Of course everybody must use the SSD as the development storage! (grin)
(my point: why should there be a norm on how an SSD needs to be used?)
https://www.youtube.com/watch?v=aoFiw2jMy-0 https://soylentnews.org/~MichaelDavidCrawford
(Score: 2) by takyon on Friday March 15 2019, @02:27AM (3 children)
One of the biggest data hogs is high resolution or raw video. Why would you store it on an SSD when you can store it on a cheaper/TB SSD? You don't need good random performance/IOPS when loading or copying the video, it's all sequential. Sure, you can get better sequential performance on a high-end SSD, but the perf would exceed the read speed necessary to play back the video in real time.
Applications and games will perform better with an SSD, so those should go into primary SSD storage. Photos and videos can be left on an untouched HDD for years without data loss, but SSDs only need to preserve data for 1 year when unpowered.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 0) by Anonymous Coward on Friday March 15 2019, @02:51AM (1 child)
http://www.linux-magazine.com/Issues/2016/187/Caching-with-an-SSD [linux-magazine.com]
https://github.com/lanconnected/EnhanceIO [github.com]
(Score: 2) by takyon on Friday March 15 2019, @02:58AM
If you don't want to pick and choose which data should go where, maybe you want to use a hybrid hard drive (SSHD).
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2) by c0lo on Friday March 15 2019, @02:54AM
Hog as in what? Size? I don't care about the size, I care about the speed.
As a developer, I care not about the load time of an application (can use one compiler process against many source files at once) and I don't produce large files, but I produce zillions of them.
I'd hate to pay thousand extra bucks for the 'special edition developer laptop' which differs from the 'consumer laptop' only by which of the SSD/HDD is used for apps and which for the workspace.
https://www.youtube.com/watch?v=aoFiw2jMy-0 https://soylentnews.org/~MichaelDavidCrawford
(Score: 2) by HiThere on Friday March 15 2019, @03:47AM (1 child)
You're making assumptions about data use. In one of my applications, the data needs to be on faster (secondary) store than the program. (The program fits in RAM when loaded, and if that takes a few seconds, who cares. The data is accessed in random chunks repeatedly and doesn't fit in RAM.)
OTOH, I still don't really trust SSDs because of random file system erases that happened in earlier years. Perhaps they've fixed that problem, but the last time I was told it was fixed a couple of other people chimed in that "No it wasn't". I understand that it's faster, but unreliability is a trade-off I'm not willing to make.
Javascript is what you use to allow unknown third parties to run software you have no idea about on your computer.
(Score: 2) by takyon on Friday March 15 2019, @03:56AM
Such problems are to be blamed on the memory controller, which might be used by multiple but not all brands.
If a disk controller is messed up, you could have a bad time with your HDD, and possibly a significantly higher failure rate.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2) by DannyB on Friday March 15 2019, @03:23PM
My corporate masters replace developer workstations every three years. I'm due for upgrade this summer. Six years ago, the standard config:
* fast core i7
* 32 GB RAM
* fast HDD approx 500 GB
* SSD 128 GB
Then three years ago config improved:
* bigger HDD
* SSD increased to 256 GB
So I wonder what the config will be this year?
But the IT department has this habit of configuring things so that Windows and the Apps (eg "Drive C") is the SSD.
Short answer: No!
Long answer: Nooooooooo!
Maybe that config is okay for marketing, sales and management. But give me the SSD completely empty.
Please make Drive C be the spinning rust. Leave the SSD empty. I don't care if Windows takes 30 seconds longer to boot. That's not what I do all day. I can efficiently use that limited SSD (especially when it was only 128 GB!) to make things faster that matter to me all day long! Like a database. Or where my source code is stored so that a search across all code is very fast. Or Virtual Box drives. And other things. Believe me, I can make good use of the SSD drive empty. The performance of Windows and apps from the HDD is perfectly fine and in fact quite good. Especially with 32 GB of RAM and a nice processor.
Now at home, I've never owned a Windows PC ever. My system is 1TB SSD + 128 GB SSD, and 32 GB RAM, fastest AMD before liquid cooling is required. I didn't put it together, I have a friend who does 'hardware stuff'. I've been quite happy with it.
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 1, Insightful) by Anonymous Coward on Friday March 15 2019, @01:25AM (1 child)
I'd be very interested in an SSD with 16 TB, provided that, per unit, it doesn't cost more than several decently-configured desktop PCs. These high-end, huge-capacity SSDs tend to have not-so-little gotchas like that.
If it's cheap enough I can mitigate potential data loss through multiple redundancy. But not at the astronomical prices that go with current, high-capacity SSD drives. In a commercial environment that kind of storage space for that kind of price probably just flat-out isn't worth it.
(Score: 4, Interesting) by takyon on Friday March 15 2019, @02:21AM
The prices may only become reasonable when a smaller quantity of very dense NAND chips are used. Necessitating at least the use of 3D QLC (4 bits per cell) NAND, with its inherent/current speed, endurance, and data retention issues.
Ideally, a 16 TB SSD would cost $1,600 (1 TB is $100 on sale or cheap brands), fulfilling your "doesn't cost more than several decently-configured desktop PCs" criteria. It would also be nice if larger capacities cost less per TB than smaller ones, rather than more.
NAND price is supposedly declining around 30% every 1-2 years. So you could come up with an estimate for when a particular capacity may reach a "reasonable" price. It seems that as long as MAMR/HAMR HDDs are delivered (HAMR is late by years), we'll continue to see HDDs maintain a cost/TB gap, although it could narrow somewhat. In 5 years, we could see 36-40 TB HDDs available.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 3, Insightful) by zeigerpuppy on Friday March 15 2019, @02:58AM (1 child)
When we're talking about enterprise storage, larger is not always better. Generally drives are ganged together in groups. More drives generally translates to better performance but higher cost. An array of 8 drives, for instance, may have 2 drive redundancy. Let's say that two drives fail and the controller/software needs to rebuild onto a new drive(s). The rebuild time is then a function of drive write speed x capacity + controller overhead. While the rebuild is happening, the array will be slower (usually controlled by an adaptive algorithm to handle build time vs access speed). A significant issue is that another drive can fail during a rebuild as the array drives need to do a full read of the whole drive, a high load scenario (assuming no write levelling). Therefore, single drive size increases rebuild time and risk of failure during rebuild (which generally means data loss). For 3.5" SAS drives, larger than about 4TB is already risky. For SSDs, it bery much depends on issues like spare blocks and total drive write speed, but 16TB sounds about as big as I'd go with current speeds.
(Score: 2) by legont on Friday March 15 2019, @03:59AM
Based on my experience, both consumer at at home and enterprise in the office, reliability picked at 250G spinning drives. Perhaps 500G were almost the same. From that point on it was all downturn.
"Wealth is the relentless enemy of understanding" - John Kenneth Galbraith.
(Score: 2) by mendax on Friday March 15 2019, @03:07AM
If no one wants to love 16 TB in one go, why would anyone want a 20 TB or larger hard drive composed of spinning rust, something that is more likely to fail. This makes no sense.
It's really quite a simple choice: Life, Death, or Los Angeles.
(Score: 1, Funny) by Anonymous Coward on Friday March 15 2019, @03:18AM (1 child)
the nsa backs up my data. i'm good.
(Score: 2) by DannyB on Friday March 15 2019, @03:27PM
Expert advice: a raid array of SD cards. For reliability! All of those SD cards connected to the raspberry pi by a single USB connector for convenience. Then make it into a storage and media server.
(do I need an /s tag here?)
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 4, Insightful) by ledow on Friday March 15 2019, @10:00AM (4 children)
Cost is the only factor here.
If 32Tb drives were cheap enough that you could have two, you'd do that.
16Tb is likely where 2 x 8Tb costs much less than 1 x 16Tb, which is why they don't want them. They'd rather get 8Tb of slightly-more-redundant storage.
And with enterprise-level SSDs, you can get into SERIOUS money very quickly.
Personally, I'd buy 1Tb SSDs till the cows come home if I could afford to. Fact that is £200 for a drive is kind of my upper limit, so I do have one in my two-drive laptop, but the second has been on my Amazon wishlist unjustifiable for years now. The initial drive literally doubled the lifespan of my laptop. Another one... wouldn't... it's just more storage. But I wouldn't buy another hard drive any more, and I wouldn't replace the other hard drive with an SSD until they are sensible prices.
Honestly, we need to just stop making consumer HDDs and pour the manufacturing and clout into making more and cheaper SSDs instead.
Most SSDs, internally, are nothing more than a non-airtight 2.5" HDD casing (you can usually just pry them open with your fingers or a flat blade) around a tiny board with a couple of chips on it. They scale really well... my Samsung 850 is 1Tb but the casing could easily fit 4-8 of those boards into the same space. They just don't.
Get the prices down, by making and selling en-masse, ditch stupid HDDs, and start making your SSDs the same prices as HDDs. I am literally holding off all storage purchases except the essentials for that moment they become more affordable.
(Score: 2) by takyon on Friday March 15 2019, @12:55PM
1 Tb = 125 GB. You mean 1 TB.
Consumer HDDs are already an afterthought. When you see a shiny new 16 or 18 TB HDD announcement, the product will be used by datacenters first, with perhaps up to a year before the same capacity point reaches consumers.
However, as long as there is a $/TB gap between HDDs and SSDs, HDDs can still survive as a category. Heck, even if there was no gap, HDDs could still have a place due to superior aspects such as data retention. They would just compete with magnetic tape instead.
They are about $0.10/GB in the U.S., at least after discounts. This is still a far cry from the $0.03/GB that HDDs paused at several years ago, but it means that 1 TB SSDs are very affordable. 1 TB of primary storage should be a good start for most people, and large amounts of data can be placed on HDDs. Apparently 8 TB can be obtained for about $140, and 10 TB for $160. So now HDDs are hanging out around $0.016/GB. ~6x gap instead of ~10x.
There are a number of options to increase NAND/SSD density, such as adding more layers, shrinking the process node, string stacking, stacking more packages, and even doubling the bits per cell to 8. HDDs have microwave/heat assisted magnetic recording, bit patterned media, shingles, two dimensional magnetic recording, and some other tricks. A gap could persist.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2) by Freeman on Friday March 15 2019, @03:15PM
I own one of these bad boys and a 512GB same brand. They're both doing quite well. Mushkin Enhanced Reactor 2.5" 1TB SATA III MLC Internal Solid State Drive (SSD) MKNSSDRE1TB - Sale Price $135. https://www.newegg.com/Product/Product.aspx?Item=N82E16820226596&Description=1tb%20ssd&cm_re=1tb_ssd-_-20-226-596-_-Product [newegg.com]
Joshua 1:9 "Be strong and of a good courage; be not afraid, neither be thou dismayed: for the Lord thy God is with thee"
(Score: 2) by ledow on Monday March 18 2019, @07:45PM (1 child)
No, if I'm talking about storage, I ALWAYS mean bytes. Every time.
If I'm talking about stupid categorisations of speeds, then I might be forced to refer to bits as a sales marketing tactic.
Not even in the days of floppy disks did we ever refer to anything storage in terms of bits, no matter what sized B symbol we used. 1.44Mb.
There is not a single reason to ever refer to groupings of individual bits in the millions like that. Especially give that every modern system barely even uses anything close 8-bit-bytes as their base unit any more.
(Score: 2) by takyon on Monday March 18 2019, @07:51PM
You replied to the wrong post, but I saw it anyway.
1.44 Mb = 0.18 MB. We know you mean 1.44 MB simply because that is a well-known floppy disk capacity [wikipedia.org]. Use the correct units.
Bits per second are widely used to refer to bandwidth speeds. Use the correct units.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 3, Touché) by DannyB on Friday March 15 2019, @03:48PM
If nobody wants SSDs bigger than 16 TB but bigger HDDs are okay, then could the objection be overcome if someone would make SSDs that are spun while in operation?
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 2, Interesting) by jlv on Friday March 15 2019, @08:56PM
"Nobody wants hard drives with over 25GB of storage" - 20 years ago
After all, how are you going to back that up?