Intel and Micron Partnership Soon To Launch 10TB SSD For Enterprise Market (hothardware.com)

MojoKid writes: Intel and Micron have been tag-teaming various storage and memory technologies and word on the web is that the fruits of that partnership is a 10-terebyte SSD that's right around the corner. The largest SSD in Intel's stable at the moment is 4TB, which itself is pretty large. However, both Micron and Intel are of the opinion that typical planar NAND flash memory has gone about as far as it can go, and that 3D stacked Flash memory is the future. They've also developed a "floating gate cell" design - a first for 3D stacked memory - resulting in 256Gb multi-level cell (MLC) and 384Gb triple-level cell (TLC) die that fit inside of a standard package. The two companies are targeting gumstick-sized SSDs reaching 3.5TB and regular 2.5-inch SSDs hitting (and even surpassing) 10TB. Apparently that's about to become a reality.

Comments

Hard Drives are dying

[Archangel Michael]

I've said for a while now, that Spinning Hard Drives are dying breed. This is just another nail in the coffin, as SSD sizes start to surpass traditional HDD. The last remaining bit that HDDs have over SSD is cost per MB. However if you include OTHER costs associated with HDDs (Watts per drive) even those advantages shrink (or go away).

IMHO once these higher density SSD drives arrive, there will be little or nothing for me to recommend standard HDD, for any application. None. There is barely any reason to have spinning drives right now.

Re:Hard Drives are dying

[110010001000]

Well one reason to keep hard drives around is because their capacity can be measured in terabytes, instead of SSD's which are apparently measured in terebytes.

Re:Hard Drives are dying

[swb]

I think the enterprise storage vendors *want* to hang onto HDDs as long as possible because it lets them keep marking up SSDs to stratospheric heights, in addition to charging a whole bunch extra for magic tiering/caching systems so the 6 SSDs you can afford to put into the thing will actually have a chance of getting used.

I also think they're somewhat scared of the evidence of greater durability that SSDs seem to have because a big part of their justification for increased cost for their enterprise SSDs involves using SLC flash.

Now, to be fair, based on what I've seen the vendors are right that you only "need" a subset of your data to live on flash because the rest of your data is a lot colder, and yes, SLC flash has a lot more write durability.

That being said, if MLC/TLC flash is generally more durable than we've seen (I've seen an endurance test that put 6+ PB on an 850 Pro) *and* fairly inexpensive, then even if it has double or triple the failure rate of spinning rust the value proposition still leans towards all flash storage because of the huge increase in performance across an array, decreased power consumption, need for fewer hot spares or parity disks due to faster rebuild times.

Overall storage system prices could be lower, too, because all that caching/tiering management takes real CPU/RAM, resulting in more expensive controller hardware with more complex and costly software.

Re:Hard Drives are dying

[Archangel Michael]

For example, if you're going to be writing to a particular sector over and over and over, SSD wears out. You no doubt will bring up that all the wear evening that SSDs use but remember those only work if the drive isn't near capacity.

For example, your example makes no sense. Wear Leveling does change it so that you CAN'T do what your initial postulation suggested. AND if you're at "near capacity" AND doing that many "writes", might I suggest that you're using the wrong size. EVEN if you were writing video streams, you'd be better off using larger capacity.

Under almost no conditions, short of "archiving" would standard HDDs perform better. And having pulled an archival HDD out of storage to have it NOT function (bearings seized do to not being used) I can assure you that even THAT application has its own problems. My suggestion is to have your data archived on a system that is always on and monitored, rather than "guessing" that it works and finding out that it doesn't.

Re:Hard Drives are dying

[Anonymous Coward]

Disclosure: I work in this field but I won't tell you who I work for.

Regarding endurance, the myth that HDDs have greater endurance than flash (even MLC and TLC) is just that - a myth. Every HDD will die eventually, frequently when they have to be turned off and won't start up again because all of the grease has been spun out of the motor bearings.

See this: https://www.micron.com/about/blogs/2016/february/the-myth-of-hdd-endurance

SSD density (at 4TB in 2.5" x 7mm or 9mm high) is already more dense per square inch than HDDs (10TB in 3.5" x 1" high). Power consumption is much lower too. Combined with the fact that HDDs today are slower per megabyte than ever (10TB drives spin at the same 7200 RPM that 120GB drives did), the near future holds the end of the HDD in the majority of applications.

That said, we have been calling for the death of the mainframe for years, as well as the death of tape for years, and neither have died. HDDs will continue to have a place in the world, but they will be by and large replaced with flash. In the enterprise space, changes in software also mean that most of this flash will be server attached rather than part of a SAN.

Re:Hard Drives are dying

[Jeremi]

Well one reason to keep hard drives around is because their capacity can be measured in terabytes, instead of SSD's which are apparently measured in terabytes.

The Terebyte is a common unit of measurement for Imperial bytes.

Imperial bytes are similar to standard bytes, except that they were invented before the widespread adoption of Arabic numbers, so instead of storing the bits as 0's and 1's, they store them as I's and II's.

Re:Hard Drives are dying

[dgatwood]

IMHO once these higher density SSD drives arrive, there will be little or nothing for me to recommend standard HDD, for any application. None. There is barely any reason to have spinning drives right now.

Except that we're nowhere near that point, at least from my perspective. A 1 TB laptop SSD starts at $240. $150 will buy you 3 TB of laptop-sized spinning storage (as long as your computer can handle a 15mm drive, otherwise 2 TB for $95). More importantly, no amount of money will buy you a laptop-sized 3 TB flash drive. (You can, however, buy them in a PCIe form factor for about $30,000.)

That last part is a major problem. Everybody is building flash drives to hit price points, ignoring the fact that flash drive capacity is still basically flailing around at 2010 hard drive capacities. In a decade, we've gone from flash drives being a decade behind hard drives capacity-wise to being six years behind. And we've gone from being 10x the price per TB to only 6x the price per TB. At this rate, the break-even point on capacity won't be until probably at least 2030, and maybe 2040, and the break-even point in cost-per terabyte is probably at least a decade away, too.

Of course, ten years ago, I'd have expected us to have passed the break-even point already. As best I can tell, the main reason we haven't is that the rise of smartphones has put a tremendous strain on the flash manufacturing industry, and they still haven't caught up. I'm starting to wonder if they ever will. If and when they actually do, then I would agree that spinning drives should go away. I'm not counting on that happening any time soon, though. I hope I'm wrong.

Re:Hard Drives are dying

[swb]

That said, we have been calling for the death of the mainframe for years, as well as the death of tape for years, and neither have died. HDDs will continue to have a place in the world, but they will be by and large replaced with flash. In the enterprise space, changes in software also mean that most of this flash will be server attached rather than part of a SAN.

I think the death of tape hasn't happened because there isn't a functional replacement for it for high capacity, long-term archiving. HDDs don't work well in changers, are more fragile and I don't think anyone trusts their powered off shelf life.

I know that clustered/distributed/server-local storage is becoming a competitor to centralized SAN, but I think it will be something of a limited market. Virtualization and CPU improvements have cut node counts significantly, making it harder to obtain redundant node counts necessary for this to see a lot of adoption.

Even the vendors with decent products now charge so much for licensing that they're not remotely competitive on pricing. I saw a price analysis of VMware vSAN that put it more expensive by 2-3x over a conventional SAN. MS Storage Spaces isn't really flexible enough yet although it's clear MS wants it to go this way, but it doesn't seem like it will be there or agnostic enough for heterogeneous workloads for years.

Re:Is this based on 3D Xpoint?

[Areyoukiddingme]

The new Intel/Micron "flash successor" that's supposed to be faster and more durable?

It's not. Both from this announcement and from the original announcement covered here on Slashdot just under a year ago, we see that it's MLC and TLC nand flash. Multilayer (a.k.a. 3D! Now with more Ds!) rather than single layer, but otherwise still bog standard nand flash. Evidently it took a while to get the yields up. Looks like they intend to crater the price per gigabyte of flash-based storage while simultaneously offering up XPoint as the (higher priced) upgrade. And it sounds like Samsung anticipated them doing exactly that, and is working to unload their single layer inventory as fast as they can.

Re:Almost there...

[roc97007]

> It might even suck so bad that no-one gives a shit about $/GB any longer, for 95% of all applications. Perhaps, minus all those cat pictures, we're almost there already.

Right. At some point it's enough for most people. I repair laptops as a side business for non-computer-savvy who have gotten fed up with offshore "support", and one thing I've noticed is that most people don't even begin to touch the capacity of the original drive. I on the other hand, as a photographer, can't get enough storage (my current machine has five terabytes -- one two and one three -- and is full up) but the average user couldn't fill up a 128 GB drive with cat photos over the life of the machine.

There are exceptions of course. A friend wants to double his laptop capacity and switch to SSD when the price comes down a little more. But I suspect it's for pr0n.

Re:Terebyte?

[110010001000]

I agree if they don't stop making these tereble mistakes I am going to leave!

Re:Problems with SSD storage

[m.dillon]

That standard is for a flash cell that is at its wear-limit (basically at its end-of-life). A brand new flash cell that has only been written a few times has 10x the shelf life.

Those numbers are also quite misleading, because in addition to the above, a powered SSD will rewrite cells when their data becomes weak. So data retention for a powered SSD is going to be a very long time. Since SSD flash cell life is based on write activity, if you don't wear it out from writing your SSD to its limit and you leave it powered most of the time, it will retain your data and probably last a very long time (well in excess of 10 years, probably in excess of 30 years, possibly even longer).

SSDs don't eat very much power, a data warehouse would be workable.

HDDs on the other-hand wear out whether they are powered or not. Corrosion, stress (even if not doing anything), lubricant issues, and any number of other factors. A HDD on a shelf isn't going to last even 5 years in any reliable sense. You might be able to recover the data from the magnetic media, but the expense would become stratospheric if you have more than one drive to recover.

-Matt

Re:Target market

[guruevi]

Anyone these days. 10TB isn't all that much anymore. If you need 200TB in the enterprise world today, you can get 4TB reliably (6 and 8 exist but are double and quadruple the price respectively and require heavy tradeoffs) so you need ~110-120 drives (RAID10+spares) and that's for 3.5", 15W.

Enterprise SSD's typically scale evenly with size so I expect these to cost ~5k each, 45 of these would do the job, at 0.2W and 2.5", those things save you first year in both power and space.