Slashdot Mirror
Microwave Tech Could Produce 40TB Hard Drives In the Near Future (gizmodo.com)
Western Digital has announced a potential game changer that promises to expand the limits of traditional HDDs to up to 40TBs using a microwave-based write head, and the company says it will be able to the public in 2019. Gizmodo reports: Western Digital's new approach, microwave-assisted magnetic recording (MAMR), can utilize the company's existing production chain to cram a lot more storage onto a 3.5-inch disk. In a technical overview, Western Digital says it has managed to overcome the biggest issue with traditional HDD drive storage -- the size of the write head. These days, an average hard drive maxes out in the 10-14TB range. But by integrating a new write head, "a spin torque oscillator," microwaves can create the energy levels necessary for copying data within a lower magnetic field than was ever previously possible. There's a more thorough white paper for those who want to dive in. According to Western Digital, MAMR has "the capability to extend areal density gains up to 4 Terabits per square inch." By the year 2025, it hopes to be packing 40TBs into the same size drive it offers today.
C'mon, guys, can't you even be troubled to proofread the very first sentence?
Il n'y a pas de Planet B.
awhh come on, let spinning HDD's die already!
AF microscope can detect electron density around atom or molecule. Just do it, WD!
Above 1 TB only geeks and IT companies care. Just like it was case for CCD above 10 Megapixel. At 1 TB disk is "solved" problem for general audience and things that matter now are performance and durability.
839*929
Some of you are missing the ramifications of this. Even though this is magnetic media it will drive down the cost of cloud storage. Right now it is cheap, but not that cheap. This could make is feasible for everyone to store all of their data in cloud for pennies a year....encrypted of course.
I object to power without constructive purpose. --Spock
Just imagine a 40TB HDD starting developing bad sectors... and then RAID arrays crashing... I really can't handle even thinking about it. I think that the use of tape backup systems for those pesky 10-14 TB disks will soon become mandatory in every home.
Give them a break. Every so often everyone accidentally a whole word.
I can't believe they're still working on mechanical hard drives.
SSD's have them beat in every way.
Put your money into that. Make THAT better. faster.
First post!!
One very basic reason the HD capacity has been held hostage at 10-14TB range is because of the disc real estate of the 3-1/2 inch format disc is much lesser than that of the old 5-1/4 inch format disc
For laptop or other mobile device, of course, stick with the smaller size 3-1/2 inch format disc, but for crying out loud, for servers which demands multiple peta-byte of disk-storage, give us the 5-1/4 inch format disc and we can at least HALF the number of physical HD we are force to use now
How much porn can I get on that?
Curious to know - prices seem to stagnate lately, drops have really slowed of late.
Picked up 5TB disks for $200 US nearly 3 years ago. There's occassionally deals which beat that but overall, its still a normal price sadly.
Also, while at it, do reintroduce 5 inch floppy disks.
Despite all you shitbag corporate shills repeating the same fvcking lies (current product is good enough), it's more and more obvious how inadequate it all fvcking is.
Petabyte hard drives, terabyte flash drives, CPUs 100ghz, and 16k displays - MVTHERFVCKER!!!
Computing is slower than fvcking shit, storage is not even bareable, and no fvcking reason (except corporate greed) why we can't have 100GHZ internet speeds.
And 40TB... I'll fill that mofo in less than a year with pr0n alone.
There is NO SUCH THING as too fast or too large in computing. And being stuck at essentially year 2000 levels (plus one miserly step) just fvcking pisses me off. FVCK ALL OF YOU.
They could make 100TB drives and most consumers wouldn't care anymore. Local file storage and hoarding has been mostly replaced by online and streaming services.
I wish they would focus on faster interface speeds, better firmware, much better caching, new algorithms, processing power and new materials for the most important mechanical parts, like the r/w assembly. Need to get the response time to 1ms. I wonder if they have looked in to carbon fiber...
The bigger news here is that 2025 is the near future.
This is an impressive density, but my question is whether drive performance increases and what kind of bus will the drive have?
SATA 3.2 claims 16 GBits/second but my guess from glancing at the other 3.2 features would indicate this is mostly a flash-oriented spec that would be tied to M.2 slots, and that the ordinary SATA slots would still be SATA-3 @ 3 Gbps.
Without a bus and drive combination capable of moving 40 TB in a reasonable amount of time (4-6 hours), these drives will be a novelty and not useful in any storage system without double or triple parity redundancy. The rebuild times without drive speed improvements and a fast bus would make them just time bombs.
How long will it actually take to write a full 40 terabytes of data? How long will it take to sequentially read all 40 terabytes? One of the reasons MapReduce is so popular is because traditional hard drives have not kept pace with memory and cpu speed increases. A 15K SAS hard drive can sequentially read less than 200 MB/sec. It would take days.
This might be useful for certain backup purposes. However, Samsung has already released a 128 terabyte SSD prototype. Obviously there is a huge price difference, but how useful is a hard drive that large if it takes forever to read all of its data? I think traditional hard drives are quickly on their way to becoming a backup/archival medium to complement tape. Live data will probably start to live on SSDs more frequently.
It's the internet gremlins! They a word and then poop them out in some other random part of the eat sentence. Gremlins I tells ya!
As always, the drops will come when the new tech arrives. 10TB drives are about half the price (~$350) now as they were when they came out (~$6-800). A 5TB drive is now at the $100-150 price point so it's dropped by 30-50%.
Custom electronics and digital signage for your business: www.evcircuits.com
Those high capacity HDDs are currently helium filled. With MAMR, one day they won't need to. Then they will drop prices.
There are various other issues why 5-1/4â donâ(TM)t work, it has to do with physics. I think the last 5-1/4â I used were an IBM DeskStar series on a SCSI bus and they had a rotational speed of 3600rpm.
Custom electronics and digital signage for your business: www.evcircuits.com
So about a 3x capacity in 7 years?
Doesn't sound particularly ambitious, nor 'in the near future'.
Toshiba X300 6TB for $169
https://www.amazon.com/Toshiba-Desktop-7200rpm-Internal-Drive/dp/B013JPLKJC/
Rsync can do pretty much whatever you want regarding symlinks. There are three different command line options. Since symlinks may point outside of the directory you're copying, and may be either relative paths or absolute paths, the "right" behavior is situation dependent. Rsync lets you choose what is right for your situation.
Once you get a few million files on these things, you can't find anything in a reasonable time frame. A 40TB HDD will hold about 40 million files if the average file size is about 1MB. Sure there are lots of really big video files, but there are lots of little tiny ones too, so an average of 1MB is very reasonable. Once you have 40 million files trying to find those 10 JPG files you took of your son's birthday party can take an hour or more. Why is that? File systems were invented a long time ago and have had only minor architectural changes since then. Searching still requires a folder traversal and examining the extension of each file name. Meta-data structures are too big (NTFS is the worst at 4K for each file record). Just reading in the file table in Windows is 160GB if you have 40 million files. If you don't want to read it again every time you search, then you need 160GB of RAM to cache it in memory. You can make things faster if you let Windows Search or Spotlight on OSX index your whole drive (basically read all the metadata and put it in a separate database), but that can take hours already. It might take days with a 40TB drive.
I am working on a new data management system that will easily handle 100 million files. You can load the whole meta-data table for that many files into just 6.4GB of RAM from a cold boot in about 30 seconds and then every query (e.g. find all my photos) only takes a second or two.
Larger drives means fewer servers per terabyte. Fewer servers means all costs drop. The dollar cost of the bare drives themselves is a fairly small portion of the overall costs, so *cheaper* drives don't make a huge difference, but *bigger* drives make a big difference.
> Further, I'd argue that I doubt most cloud storage companies list "disk storage" as their primary expense
40TB drives would mean the cloud providers need 90% fewer servers. A tenth as many servers means the datacenter can be 1/10th the size. 90% fewer servers means 90% less tech time of employees going around swapping out bad drives, running cables to new servers, etc. Basically ALL the costs other than marketing and internet bandwidth are a function of the number of drives.
Looking at it another way, if a provider is currently using 4TB drives, switching to 40TB would mean they could have 10 times as many customers in the same datacenter, on the same servers, for the same cost. So yes, most of their costs are directly proportional to how many disks they have spinning.
> I'd similarly argue that even if being able to quadruple storage capacities per 3.5" bay did save them a bunch of money, that they would be unlikely to pass that cost along to customers.
There is in fact competition in the cloud storage market. A lot of competition. So much competition that until recently, major companies were willing to lose money on every customer in order to get more customers (they planned to start making money after the fast growth of the industry began to slow). Cloud storage companies absolutely are busting their ass to give customers the best service they can at the lowest price. They want the business, so they're offering the lowest price they can.
Amazon is of course the premium brand in the space. What they offer their customers is being the *best*, not the *cheapest*. Amazon's cloud has so many useful features added now that the documentation is well over a thousand pages. They give customers "the best bang for the buck" by constantly adding more bang, for the same buck. To that end, Amazon is spending over a billion dollars a month on R&D to constantly improve their products.
Youd've have to cut RPM which is okay in most cases (like watching porn) since SSD has taken over that niche. The basic problem is home cases are built for 3.5" and anyone who needs more space will use one of the readily available arraying schemes. It's not like 5.25" will make anything much cheaper so it's a solution crying out for a problem.
give us the 5-1/4 inch format disc and we can at least HALF the number of physical HD we are force to use now
Why stop there?
The very first hard drive, the IBM 350 RAMAC, had fifty 24-inch platters. If we went back to that form factor, with this new technology you could pack over 130,000 TB into a single drive!
Bring back laser-discs!
"There are various other issues why 5-1/4â donâ(TM)t work, it has to do with physics."
No, it has to do with material construction. We've got plenty of materials now days that could allow a 5 1/4" platter to rotate at 10K RPM without failing. My 6" trim saw runs faster than that with a less-than-1mm thick blade.
Still waiting on Serviscope_minor to wake up to fucking reality and realize that Jessica Price isn't going to fuck him.
No, even with MAMR they'll still need to evacuate the drive case and fill it with helium because you can't risk water molecules in the enclosure attenuating the microwave emissions from the heads and causing bad data writes.
Still waiting on Serviscope_minor to wake up to fucking reality and realize that Jessica Price isn't going to fuck him.
And; as a bonus, that external drive (it would pretty much have to be an external wouldn't it?) would be amazingly theft resistant! Of course, power consumption would be a bit of an issue, but I'm sure the manufacturers will promise to fix that with the next firmware update.
I need a wheelchair van for my son. Help me get the word out. https://www.gofundme.com/wheelchair-van-for-jj
You've a very good point that "rsync" provides options to handle symlinks differently. But those options are aimed at the correct replication of a symlink on the source side. The transfer of content on top of an existing symlink normally breaks the symlink on the receiving side, unless it is a matching symlink.
If you've seen a way to get it to consistently transfer plain file content from the resource side, on top of an existing symlink, leave the symlink untouched and publish the content to the target of the symlink, I'd be curious to see the combination of options you're using.
Were these the size of a top load washing machine? Maybe they got the size down to an under counter, home dish washer.
In a time of universal deceit, telling the truth is a revolutionary act. George Orwell
40 TB, worser or better MTBF?
At 40Tb / square inch, how many molecules are there per bit? Or, how close are we to the absolute maximum density or 1 molecule / bit?
when everything matches up
Sigh.
When you're trying to recover a spinning-rust ZFS volume, what you really care about is independent head-servos per terabyte, if you want your mean-time-to-recover to be a smaller number than your mean-time-to-cascading-failure.
Sure, you could build a viable ZFS storage system using multi-petabyte hard drives, so long as you can perform a complete drive read (semi-sequential) in under 24 hours (for a 24 PB drive, that works out to a sustained per-drive semi-sequential bandwidth of 280 GB/s).
And then you'd need a really stiff platter axle, and fluid dynamic bearings able to support a 5 kg platter assembly, spinning extremely fast, with no wobble whatsoever, unmaintained, for years and years.
How many Hot Pockets Snack Bytes can it heat up? Programmers get hungry ya know.
It must have been something you assimilated. . . .
It's an older meme, but it checks out.
No, the RAMAC was more of a walk-in freezer. You're thinking of the IBM 2311 a few years later, which was a top-loading washing machine that gave you over seven megabytes of storage. Nobody knew what to do with all that capacity.
... ever done any traveling? Data-roaming charges can bankrupt you.... Ever traveled to the country-side where you have a really crappy connection?
I have music in FLAC format that i cannot stream online... I have quite a few pictures taken over the last 25 years that i want to save... I have old video's i want to save... I have a bunch of movies ripped to disk so i can do playback without having to switch disks.....
Sure there are many things that can be streamed, but all those streaming-services continuously remove and add things to their service so you never know what you will have access to..
I do remember having stored a 5.25" Quantum Bigfoot somewhere. It's not anymore in service, but it operated for about a decade in my room in the student dorm I lived, around the turn of the millennium... Slackware based 486 router PC (I was the unofficial local ISP at that time) with an experimental cable modem connection, 2 NE2000 compatible ISA NICs (coax was more robust when squeezed between the door and post of the dorm rooms...) and a single 'luxurious' 3Com 3c509 with 10base-T for the room-local net... The disk did 4000rpm. Couple of GB in size, IDE interface.
Jeff Frick interviews Brendan Collins on MAMR — 12 October 2017
15 seconds logo rotation.
30 seconds lip-gloss application.
30 seconds applied lip gloss.
At 2 m mark there's a flaccid PMR confession.
2m30 Finally, a useful question. Answer mentions head process "Damascene".
3m30 three enabling technologies for last three to four years: helium, microactuation, Damascene process.
Then, effectively "this new thing today makes things better blah blah blah track density blah blah blah linear density blah blah blah".
4m50 guest decries "host-side modifications" and notes that HARM requires wear-levelling, which introduces host-side modifications.
Then a truly inane comment from Frick, "heat is bad, isn't it, for everything in electronics?", which causes Collins to chuckle nervously, before he vigorously redirects, managing to instantly step in it himself noting "an order of magnitude heat difference".
Hmmm. HAMR runs circa 900 degrees K, so that would place MAMR around 90 degrees K (we'll exclude 9000 degrees K). If true, could require innovations involving liquid nitrogen. I think he meant "differential heat change", but he realizes he's talking to a Wharton MBA, so he kindly lays up.
6m30 MBA-style "more, more, more" masturbation. Ack! Thppt! (Bill the Cat saying "gag me with a spoon, and I mean right now".)
This is not a Wharton recruitment video.
Does your trim saw run with less than a micrometer of flex on the blade and have less than 10W power consumption? 10 and 15k RPM disks exist in 2.5" and 3.5" but they become significantly more expensive to produce, increasing the power, thus heat production to the point that SSD is simply cheaper. SSD is already becoming cheap enough for many datacenter purposes when you're taking into account power costs.
Custom electronics and digital signage for your business: www.evcircuits.com
I was extremely snarky in another post about an MBA-level chit chat on YouTube that revealed basically nothing.
Now I've tracked down a 1.5 hour technical talk featuring Mike Cordano, Dave Tang, Janet George, Brendan Collins, and Jimmy Zhu.
Technology of the Future: Western Digital Announces MAMR for Next Generation HDDs — 12 October 2017
The first 6m30 are disposable, that's as far as I made it so far.
This was via a failure forum where Seagate employees gather to trash Seagate management. Wow, what a potent reminder of the Y2K post-bubble apocalypse.
> Who needs local storage beyond the OS and apps in this day and age when anyone can store their data on the cloud
Anyone that wants it in a timely fashion.
Cloud storage absolutely SUCKS for performance because performance of the network sucks. That's not even getting into the reliability, portability, and cost issues.
Even local gigibit NAS storage is hopelessly out of date in terms of performance.
A Pirate and a Puritan look the same on a balance sheet.
Most personal storage is going to ssd. End of story.
This only matters to the cloud guys, big brother, and the investors that they're trying to keep from jumping ship.
So the source is NOT a symlink. You want the destination to be a symlink, and you want it to copy from src/a/file to destination/b/somewhere/ file ? So the file contents up up somewhere totally different than where they are on the source, based on a previously existing symlink on the destination?
If I'm understanding you correctly, you can probably achieve the same goal by trading the symlink for either a hard link or a bind mount. If the symlink points to a directory, use a bind mount instead. If the symlink points to a plain file, consider a hard link.
Came here to say the same thing. I can picture BeauHD asking for a raise and his boss saying, "lol, what? You know you're really shitty and lazy at your job? Grade 8 students put in more effort than you and they don't get paid".
How about someone who does not want all those cloud storage providers having ultimate control over all their data? Even if you encrypt everything so that they don't spy on you, sell your private information to the highest bidder, or turn it over to the government; they could still destroy everything or simply block your access to it if you refuse to pay their new 'access' fee.
Doesn't time just fly? Whippersnappers these days have no idea how valuable time is.
military been using this tech in space for decades. next up they'll get rid of the spinning disks and switch to holography using interferometry to read and write to the matter.
https://www.trumpsweapon.com/
"Does your trim saw run with less than a micrometer of flex on the blade and have less than 10W power consumption?"
Yes, it has to for high-precision lapidary work, especially on precious opal, in fact I can put my entire weight on the blade and it barely flexes. Power consumption, 10W every hour nonstop? Yes, actually less than that since it runs on 5V 400 mA.
Welcome to like 1980, old timer. This sort of tech and alloy has been around for DECADES, you're only now catching up to what us faggy jewelry types have had access to for forever, your shit companies just didn't want to pay for the tech back then.
Still waiting on Serviscope_minor to wake up to fucking reality and realize that Jessica Price isn't going to fuck him.
hopefully SSDs-like drives will reach at least 40 TB (so that we can say "no thanks" to another heavy, noisy, power greedy and slow technology).
Slashdot, fix the reply notifications... You won't get away with it...
I'm afraid that rsync normally deletes the local file and replaces it with the new file, doing the replacment either bore the complete delivery (for certain options) or after completed delivery to a temp file. That breaks the hard link. Rsycing hardlinks is quite tricky if the hardlink exists on the _target_ filesystem, and not on the source filesystem, and does not happen if the hardlink leads outside the target rsync directory. "bind mounts" do not work for individual files in Linux filesystems, only for directories. So keeping individual symlinked files that may point elsewhere, to other attached filesystems, can be quite awkward.
I'm afraid that this kind of confusion is a common one. Developers and less experienced systems personnel believe that their casual familiarity with a tool applies to edge cases. It's when they have to actually do the work and deal with edge cases that they learn, sometimes quite painfully, that their original simple model can prove very complex to maintain, even proving incompatible with their other working tools.
I do apologize if I seem condescending about this: My experience with rsync and its limitations is hard-won, and its subtleties can be confusing.
> I'm afraid that rsync normally deletes the local file and replaces it with the new file
---inplace
>. "bind mounts" do not work for individual files in Linux filesystems
That's what I said. I said "if you're symlinking to a directory, consider bind mounts"
> and does not happen if the hardlink leads outside the target rsync directory.
Hard links don't lead to any directory. Hard links (aka file names) lead to disk blocks. We call them "hard links" when two or more different file names happen to lead to the same disk blocks, but neither name refers to any other. They are just regular file names, pointing to blocks on disk.
> Rsycing hardlinks is quite tricky
Which is why we provide options such as:
--copy-dest
--link-dest
--backup
---inplace
You're also allowed to run rsync twice, with different options, in a two-step process (that's a big giant hint right there)
> less experienced systems personnel believe that their casual familiarity with a tool applies to edge cases. It's when they have to actually do the work and deal with edge cases that they learn
People who didn't even read the man page carefully believe that their casual familiarity with the tool applies to edge cases. It's when they actually do the code in rsync, writing options that handle edge cases, that they learn how it actually works.
> I do apologize if I seem condescending about this: My experience with rsync and its limitations is hard-won, and its subtleties can be confusing.
I do apologize if I seem condescending about this: My experience with rsync is from stepping through the code and contributing improvements; the subtleties of my code can be confusing. (Especially if you don't read the entire man page)
If it was and they could get 130k TB into it, I'd find a way to live with it.
"--inplace" breaks symlinks and hardlinks. I just tested it under CygWin and under a current Linux.
"--copy-dest" simply replicates unchanged files. That is not part of the "copying content from elsewhere on toop of symlinks and making sure the conent is copied to the target of the symlink that I was trying to explain.
"--backup" doesn't help the issue. The backup exists and you can derive the old symlink to script around the issue. You can also do a "--dry-run" and parse that to deduce what needs to be synced locally and what needs to be copied elsewhere to copy correctly to the symlink target.
"--link-dest" is entirely useless for the original suggested purpose of "On an operating system with working symlinks, you can install part of a game on ssd, and part on HDD. "
I'mn afraid that ensuring that a local symlink in the target directory gets replication copied to the target of the symlink cannot be done well by a simple symlink rsync command. It requires separate tuning, _outside_ the programming of the rsync command line options.
Also, at least according to the current "ln" manual page, you *can* hardlink directories with the "-d" option, depending on the filesystem. I admit that it seems to be disabled on the first Linux ext4 filesystems I tried it on, and under CygWin accessible NTFS. But it used to be a much more commonly supported option. I'm not clear whether it's been entirely discarded.
Goodness, my typing was horrible in my response. It detracted. The core of the message should stand, but I do need some sleep. If the danger of relying on rsync to consistently copy non-symlinked source content to local symlinked content, and expecting the changes to propagate reliably to the symlink target, is unclear, I'll revisit the issue.
Symlinks to place content on another fileystem is useful. It needs to be handled with caution.
I pretty much told you how to do it, and marked that with "(this is a big hint)". Do you want to solve your problem, do you want to know how we do it, or do you want to keep arguing that it can't be done?
Link to the white paper referenced in the article but absent from /. entry: https://wdc.app.box.com/s/2o32m71nthf84ozbz8wlwyo59uqxb6og/file/237010181169