Cannot imagine wanting under any circumstances
Hard drive technology advancement has crashed into a brick wall. Stuff like this is just not useful. The horrible warts are simply not worth the tiny increment in density.
HGST has cracked the 10TB barrier with a shingled helium drive for active disk archives. Think of it loosely as a shingled He8. Like the He8, the Ultrastar Archive Ha10 is a helium-filled drive in which platters spin with less friction and vibration than in an air-filled drive. The same physical drive enclosure can hold more …
Agreed. Stop faffing about and make reasonably-priced 4Tb SSD's or similar instead.
This kind of thing is just wasting money compared to what you could do ramping up production of modular SSD's and getting to the point where you can just put together 4 layers of chips in one drive, 10 in another, and only one in another, and sell them to anyone and everyone at a range of decent prices.
1Tb SSD's are ALMOST there. Really close. Another 10% or so off the price and I'm sold. But, no, everyone else is still messing about with hybrid, helium, spinning disks, and faster and faster SSD's instead. Don't. Just give me reasonable storage (that is, after all, the primary purpose of a drive) at SSD speed (not hybrid) at a decent price, and I'll buy five.
If you too are basing your opinion on your need for a few TB of RAIDed media for your home or small business NAS/SAN, you're coming at this from the wrong angle - this drive is not for you.
Sure, cheaper flash-based storage would be great - even better if it could be built into massive, low-cost, multi-petabyte, arrays that could be powered down for maybe long periods, without fear of data loss, to save on running costs...but as of now the financials and technicals aren't right for that, so here we are. It'll come - whether NAND flash-based, or some other technology - but not yet.
I can think of plenty of applications where it's useful.
To start with, anything read-mostly. Fill the disk then hardly ever change anything on it. Wonder how many YouTube / Google will be buying?
Then databases, accessing the disk directly. Storage of large objects, or aggregates of small ones that get read, updates and rewritten as a block.
Anything, once a filesystem understands the underlying technology and can aggregate writing of small objects. It's a very good match to "everything is journalled" filesystems where nothing is updated in-place. Probably still best used for filesystems that get read more than written. Archival springs to mind. So do several of the computational physics modelling projects I used to be involved with. Those folks could never get enough storage to keep a fraction of the old results that they wanted to keep.
Interesting lack of mention of anything Microsoft or Windows in that diagram!
> I can think of plenty of applications where it's useful.
So can I.
In fact, I know of one person already using a Seagate 8TB shingled drive in his MythTV box. Recordings are made to "normal" drives, then stuff he wants to keep is moved (when the system is reasonably quiet) to the shingled drive which fits more capacity in but with the already known about write performance issues.
Once moved, the recording becomes "read only" until such time as it's no longer needed and gets deleted.
So there's one application to start with - media servers. Big space, write little, read many. I could see these drives being a big hit with the "big names" serving up huge amounts of video etc on-demand.
And as someone pointed out, an ideal fit for much of what Google does. Racks and racks of servers with "read mostly" data that doesn't get changed often, and when it does get changed then it gets changed in big chunks - when whole databases of web/link/search index data gets updated.
And then there's backups. Take a chunk of data from a system, throw it onto a backup drive, job done. Mostly sequential writes of large chunks of data - just what these drives are designed for.
And some of the write shortcomings can be mitigated by system design. Eg IIRC XFS can use a separate drive for a filesystem cache - so combine one of these shingled drives (or an array of them) with a fast (but modestly sized) SSD and you could have a real performer for certain workloads.
I wouldn't want one in my desktop, or indeed most of my servers, but I can sure think of uses for them.
I think that the OP is right when they say "Hard drive technology advancement has crashed into a brick wall".
Yes, this drive is only designed for a very niche use, that's the point. It used to be that the high capacity drive technology would trickle down to more mainstream uses, but consumer drivers have been sat at 4Tb for several years now.
Spinning rust as a storage medium has pretty much reached it's limits. I'm going to go out on a limb and predict that we never see a harddrive bigger than 20Tb in a 3.5" package.
I have an 8TB Seagate shingled drive sitting on my desk right now. Definitely a consumer drive (~AU300) and definitely useful. I shoot a lot of HD video and I have no intention of losing it to a crash. Speed is perfectly adequate in that I haven't noticed my backup jobs to be significantly longer.
Spinning rust as a storage medium has pretty much reached it's limits. I'm going to go out on a limb and predict that we never see a harddrive bigger than 20Tb in a 3.5" package.
Assuming you mean a drive containing a conventional hard disk purely electromagnetic write head, you are probably right. The physics limits for that have been reached. Magnetic domains can't be pushed closer together along the tracks (they become unstable if made any smaller), and heads cannot be made any narrower (the head would have to fly much closer, and there are power density issues even if that hurdle were overcome). Stacking tens of platters isn't likely to work either (too flexible / vibration coupling issues).
On the other hand, there's HAMR which uses a focussed laser to address a narrow track. It "softens" the magnetic film by heating it, so that a wide magnetic bubble from a conventional head operated at lower write intensity can change magnetisation only of the laser-addressed track and not affect the adjacent already-written ones. I expect that can go to many 10s, maybe 100s, of TB per 3.5 inch disk if it can get out of the lab, and if there is a market for such huge drives.
There's also BPM, about which I know less.
Wait and see. It's still possible that solid-state will obsolete all magnetic disks, but not current Flash which is also close to its physics limits. 3D flash may reach an affordable TB scale. Memristor tech has the potential to surpass that. It's rather further away than HAMR.
2 drives vs one of same volume:
2 x Electronics
2 x Motors
30% more casing
More power
More platters on the one spindle means more than x2 storage vs two drives. We have had loads of different form factors before. Ultrabooks, laptops, netbooks etc need skinny drives. Archival systems can use a different chassis. A double height 3.5" HDD will fit inside many PC and cheap server cases. 18T storage should be possible.
Actually I'm fairly certain that if they could get 20x capacity by stacking a dozen or more platters, each with twice the area, in a 5.25 inch full height container, then they would. But there are good physics reasons why they can't. Such large stacks will have all sorts of extra vibration modes, and failure to tame any of these would make the whole project non-viable. There's also the extra inertia of a bigger stack of heads, increasing problems with inter-drive vibrational coupling for those who design whole storage arrays. And of course, multiplying the number of heads and platters will considerably reduce the MTBF of the assembly, to the extent that it is heads and head/platter contact issues that dominate HD failures. I'm not certain as to this being the case, but in my experience over half of disks fail "soft" with deteriorating SMART metrics and increasing bad block counts. "Instant Brick" is relatively less common, especially once drives have survived in service for a month or so.
Too bad I wont be able herd these in.
340Tb of Legal hold and financial historical data - and they want it *off* of tape since it takes too damned long to get at now. 160-180 of these in two arrays in different geo's, and a block duplication FS on top, or reasonable raid.
I could imagine repopping a pair of frames with a ZDL type appliance in the middle and taking 10 weeks to migrate off. But that would be too easy for management as it has $$$ attached to it. Meantime they want to use anything that has the room but doesn't have $$ attached. Or for that matter offsite. *sigh*