back to article Disk is dead, screeches Violin – and here's how it might happen

And in a flash, disk is dead. Well, actually, not that fast. But flash array flogger Violin Memory is convinced disk is dying in the data centre, killed by a series of NAND acronyms: 2D MLC, 3D TLC and, the final blow, 3D QLC. Is this view realistic? Even remotely realistic? It starts from a pair of observations by Violin …

  1. This post has been deleted by its author

    1. Charles 9 Silver badge

      I think it's a little more than this. LaserDisc may have had the wow factor thanks to its relation to compact discs, but it had genuine technological advantages. For one, since it was an optical medium, LaserDiscs would be left open compared to the caddys of CEDs. It also helped that LaserDisc could keep up with CED (using CLV mode--not as many bells and whistles, but it worked). But neither one took off in the home market because the VTR had one key advantage consumer optical media could not allow until around the time of DVD; It couldn't RECORD. LaserDisc's bacon got saved by the specialty market that could use the disc's more exotic features (like image storage and frame-accurate seeking, exclusive to LD CAV mode) to display interactive video.

    2. Naselus

      So something considerably superior in almost every way only killed off inferior opposition because of the 'cool' factor? I'm a little dubious of your reasoning there. Besides, the analogy doesn't hold up that well - flash gives noticeably better performance than HDD in almost all respects even for home users, which is why it's murdering HDD in those markets where cost is similar. It's not because all the cool kids are using it (since the cool kids - more or less by definition - couldn't tell you the difference between a 7.2k HDD and a 3d NAND SSD, or whether they have one or the other in any given gadget).

      As to the article itself, I don't really buy the idea that flash will kill off HDD completely, any more than HDD killed off tape completely. Every time they manage to decrease the $/GB of flash tech, HDD decrease their $/GB even further. Some times, you just need a stupid amount of space cheaply and don't really give a damn if it take half a minute to access it rather than half a second. HDDs will just take their place in the hierarchy of cost vs capacity vs speed. They won't be used for everything, but they'll go the way of tape, rather than the way of the punch card; cheap, slow, huge, rarely-accessed data storage.

      1. Anonymous Coward
        Anonymous Coward

        "Every time they manage to decrease the $/GB of flash tech, HDD decrease their $/GB even further."

        But isn't spinning rust starting to hit physical limitations, which is why they've started to resort to such exotic techniques as shingling and heat-assisted magnetic recording? At least tape is a tested technology in regards to archival-grade (>5yr) storage plus has the benefits of being a more flexible technology; need more capacity, just wind more tape and so on.

        1. This post has been deleted by its author

      2. This post has been deleted by its author

        1. Greg D

          Again flawed reasoning. Laserdisc was superior in every way except cost.

          The RIGHT format won out. Unlike the VHS/Betamax argument.

          Also happened the correct way with Bluray/HD DVD. Although it required refitting fabs to build blue lasers and new discs, and was more costly in EVERY way than HD-DVD, it was the better format and had more potential for future improvement.

          CED had literally nothing. It was a novel use of vinyl technology and from it's inception was already at the peak of how good it could get.

          1. JimmyPage Silver badge
            Stop

            The RIGHT format won out. Unlike the VHS/Betamax argument

            The reason VHS won out was that while Sony was paying engineers to (successfully) improve Betamax, JVC/Philips et al were paying salesmen to negotiate "VHS only" deals for movie rentals with the studios.

            The result was the consumer saw the VHS section of their local video rental store far outnumbered the Betamax, and bought (rented) their equipment accordingly.

            Why do you think Sony when on a massive spending spree in the late 80s and 90s buying up - wait for it - content providers ?

            1. DeepStorage

              Re: The RIGHT format won out. Unlike the VHS/Betamax argument

              No, the reason VHS one is that Sony's smaller Betamax tape held 90 minutes at top quality while a VHS tape held 120 minutes. Most feature films fit on one VHS tape, a much better deal for the studios and video stores, vs 2 for Beta.

              Sony was also quite tight with Betamax licenses while JVC let anyone that paid the fee make VHS so the RCA VHS deck was cheaper than the Sony Betamax at Macy's.

              People incorrectly use VHS/Beta as a victory of marketing over the best product. It's actually lesson that different constituencies have different definitions of best. The studios valued run time over picture quality, the customer low price and familiar brands.

              Sony bought up studios in the 80s and 90s for the same reason Comcast bought NBC more recently, controlling both content and distribution is profitable.

          2. This post has been deleted by its author

            1. ceebee

              of laser rot and things

              It is ironic that LV..despite superior performance suffers from "laser rot" (the glue holding the clear plastic coating to the aluminum disk goes opaque) which means my CEDs play fine 30 years after they were pressed but many of my LV disks of the same era are totally unplayable!

            2. Greg D

              No, I didn't downvote.

              I read it all and couldn't work out the analogy, since flash disk is so obviously superior in every way it doesn't seem to draw any parallels with CED/Laserdisc standards battle.

              Disclaimer: I was only born half way through the 80's, so I'm using anecdotes and wikipedia for my information on CED. I was also going to point out that both formats pretty much lost out to CD in the end anyway, followed by DVD.

              1. Charles 9 Silver badge

                "I was also going to point out that both formats pretty much lost out to CD in the end anyway, followed by DVD."

                No, both formats lost to the VTR because VTR had a killer feature; it could RECORD shows. When DVD came along around the turn, VCRs hung in there because they could still record. It finally pretty much saw the light when consumer DVD recorders hit the market.

      3. This post has been deleted by its author

        1. Naselus

          "But isn't spinning rust starting to hit physical limitations, which is why they've started to resort to such exotic techniques as shingling and heat-assisted magnetic recording?"

          Yes; but then, Flash has even more aggressive physical limitations due to the minimum practical die size. I've seen nothing to actually convince me that QLC will allow capacity parity with the largest potential HDDs without necessarily compromising write life. Moreover, with flash-killer techs in the pipeline and RAM-speed storage approaching market, it's questionable if Flash has the time to finish off spinning rust; 3rd-gen SSDs might not hit a reasonable level of production maturity before they too lose ground. It's important to recall that the speed at which we've gone from disk-to-flash-to-next-thing is vastly smaller than the time we're been piddling around with disks. HDDs have taken over the market over the course of what, 50 years? Flash has less than half that before it's replacement arrives, at which point production for flash will stop expanding and start decreasing (probably without actually reaching the level of HDD output in Exobytes/year).

          "So, logically, the last hold-out of spinning rust will be in MAID systems. But will flash eventually replace disk there, too? Archiving the dross from the large social networks is basically a write once read never, (WORN), application. As such, the degration of flash with multiple re-writes doesn't matter.

          When will we see Massive Array of Low-cost Flash, (MILF), systems?"

          Only if we reach the point where either a) flash manages to reach price parity with disk (unlikely for the reasons above), or b) disk manufacture ceases completely (plausible, but not within flash's economic lifetime).

          El Reg's own Chris Mellor, a man who knows his way round a SAN, outlined a largely similar argument only a few months back (http://www.theregister.co.uk/2014/12/09/no_flash_datacentre_takeover/) - he estimated a $300 BILLION investment in new foundries was required just to bring them up to producing enough to sate current HDD demand. He places that as a conservative estimate, though I'd suggest the better dedupe and reduced need for over-capacity makes it less of an issue - if we presume that flash can do all that HDD currently can using only 1/4 of the space (a generous estimate; I'd guess it's actually more like 35%-50%) we're still looking at $75 bil extra capital investment, at a point where 'NAND-killers' keep creeping out the woodwork with predicted ten-year-or-less time-to-markets (though yes, HP already predicted their NAND replacement would be available two years ago, so a pinch of salt is required.).

          1. This post has been deleted by its author

            1. Naselus

              "OK, I hear what you're saying about flash, and mostly agree, but s/flash/solid state storage, and the picture changes somewhat - I just don't believe that spinning rust is going to stick around much longer."

              I think that HDD is likely to get killed at the same time as Flash, tbh; when we get a storage solution that is as quick as RAM. That will require a big change in software architecture that will largely obsolete existing drive and memory tech altogether. It's a cataclysm model rather than a gradual change.

              1. This post has been deleted by its author

                1. Charles 9 Silver badge

                  "But in this post cataclysm world, what do we do about long term archiving? Stuff we actually want off-line? Or are we supposed to change our way of thinking and eliminate the concept of off-line storage altogether? (Please no...)"

                  In terms of archival-class storage, tape still wins. Its physical properties are better suited for the job than any other medium on the market to date. Thing is, it's become so niche that it's basically an enterprise-only solution now. In contrast to the late 90's when it least had a prosumer solution which I miss terribly; I would LOVE to have something that can keep terabytes of data safe in cold storage for about 5 years plus (basically long enough that it maintains its integrity until it's replaced with the next evolution). Hard drives are rather iffy at that length of time, and everything else is too small, not reliable enough, or both.

                  1. Naselus

                    "In terms of archival-class storage, tape still wins"

                    Currently, yeah. I do recall reading something about crystal-based storage technologies being worked on at one university or another a few years back; they were claiming a practically infinite life expectancy (measured in Thorium half-lives or some such equally preposterous unit of time), exobyte-scale capacities and very, very low costs, but glacial read/write speeds. But traditionally, new offline archive techniques haven't really been commercially viable.

                    I do think tape will probably struggle in a post-Volatile world, though, simply because it's not hugely efficient in terms of restore speeds. There's already issues with that emerging in DR work with AFAs - you just can't get the data off a 200TB+ tape quickly enough. The bottleneck until recently has been the disk rather than the deck, but once you have RAM-speed storage and 1PB+ archive storage, the problem centres on the tape deck.

                    1. Charles 9 Silver badge

                      "Currently, yeah. I do recall reading something about crystal-based storage technologies being worked on at one university or another a few years back; they were claiming a practically infinite life expectancy (measured in Thorium half-lives or some such equally preposterous unit of time), exobyte-scale capacities and very, very low costs, but glacial read/write speeds. But traditionally, new offline archive techniques haven't really been commercially viable."

                      That tech has been "around the corner" for nearly 20 years. Australians and Americans may have seen the tech once or twice on an episode of Beyond 2000. From what I can tell, the big problems with the tech has been (a) getting the writing and reading to work together in a precise and reliable manner and (b) media longevity issues that aren't immediately apparent, such as destructive reading and thermodynamic stability (as in most crystals aren't as stable over geologic time as they appear; even diamond isn't that stable).

                      http://physicsworld.com/cws/article/news/2013/jul/17/5d-superman-memory-crystal-heralds-unlimited-lifetime-data-storage

          2. Charles 9 Silver badge

            "Yes; but then, Flash has even more aggressive physical limitations due to the minimum practical die size."

            But at least Flash can still go 3D and has plenty of room in that regard. Magnetics are already 3D (both in terms of platters and in terms of perpendicular recording) so are running out of ways to cram more data.

    3. Greg D

      Not something you can compare IMO. Flash has WAY more advantages than you give it credit for, with perhaps the only drawback being longevity and reliability.

      Fact is, magnetic spindle storage has limits, and it's reached those limits in terms of data throughput. It just can't keep up with modern demands.

      Flash is cool, yes. But it's also lighter, smaller, orders of magnitude faster (and getting faster still), and less unfriendly to the environment (less materials required to build, not sure on fab process!).

      1. John H Woods Silver badge

        "... less unfriendly to the environment (less materials ..." -- Greg D.

        Two more advantages: durability (especially useful in laptops) and power consumption (useful in laptops; probably unimportant in the desktop PC; but significant in datacentres)

        1. razorfishsl

          Flash is NOT durable.... data loss potential is far higher than disk.

          One point that is not mentioned... is than not only does writing decrease the life.... so does reading.

          It also self erases faster and if the read amps go out of sync you loose everything.

          Not to mention destructive 'cascading' issues, when you erase 512 bytes, you have to erase a full 'block', the contents need to be moved to another block, if that block fails then you find another.

          This process can cascade and then you have no place to put the data you are trying to save, it is compounded by the fact that you CANNOT just overwrite a 'page' as in a normal spinning rust setup.

  2. thegreatsatan

    For a company that saw a 33% drop in revenue

    I hardly think they would be the arbiter of the future direction of storage.

  3. techlogik

    Genius

    "Flash has endurance problems, because the cells in its blocks wear out with repeated writes." Their approach is genius. Just like light bulb makers. They can make a light bulb that last 100yrs easily you could buy, but, then they can't keep making them to keep revenues alive. A continuous death keeps the process flowing and capitalism alive and well. Disk is dead. All desktops and tablets replaced in our last refresh now have SSD, speed, heat/noise operation way less and sizes are smaller. It will keep leading to smaller, less power consuming devices while improving performance. Is it the fastest technology that is possible? No, but spinning disk are going away in the near future. Rather take a chance of blocks wearing out than a ball bearing spinning for a long time that will always fail eventually itself. Not sure why anybody thinks it is any worse than moving parts concepts.

    1. razorfishsl

      Re: Genius

      Also repeated reads.

      The difference is that the failure model is very difficult to analyze, with disks you get a strange sound or other odd behaviour.

      Loosing a sector in a disk is usually a warning sign, with flash loosing a 'sector' is 'norma'l behavior

      making a catastrophic failure very difficult to track.

      If you could get your hands on the shit that these chip manufacturers DON'T allow access to and require an NDA to see, you would get a VERY different picture.

  4. Bucky 2

    Let's make this interesting

    Everybody likes to make predictions about the future. But there's no real downside to CEOs making silly "forward-looking statements."

    Let's change that.

    Have the Violin dudes set a date, and a goal. Not for themselves, but for flash.

    If the goal is not reached by that date, they must perform a dare. I'm thinking everyone in the company must shave his eyebrows off.

    1. John H Woods Silver badge

      Re: Let's make this interesting

      "If the goal is not reached by that date, they must perform a dare. I'm thinking everyone in the company must shave his eyebrows off." -- Bucky 2

      Maybe we can let the company off for being optimistic and blowing their own trumpet. But this is an excellent idea for application to Gartner and the other f̶u̶t̶u̶r̶o̶l̶o̶g̶i̶s̶t̶s̶ analysts!

  5. stucs201
    Joke

    To quote Tony P.

    "Disk is not dead! Disk is LIFE! "

    Ah, wait, that was disco. As you were.

  6. baka

    TLC and QLC suck at performance

    TLC today (non-samsung) is about 300-500 PE cycles. Thats 1/10th of MLC. I am sure QLC is going to be even worse. Yes your SSD controllers can expand the life of these, but not enough to make up the difference in cost (up to 20% cheaper) and performance (I bet QLC will be a real slug).

    Oh boy a cheap SSD that lasts 3 days in a server and is still more expensive than an HDD!

    The labor costs to keep switching these out will likely outweigh the small performance benefits.

    Also, 3D NAND has a long way to go to scratch today's MLC and TLC pricing. Its a brand new (read: expensive) process that until it is a 1/2yr into major production will be much more expensive.

    QLC, sigh.

  7. Sarah Balfour

    Initialisms, NOT acronyms…

    …unless it can be pronounced as a word, it's NOT an acronym.

    1. This post has been deleted by its author

      1. Notas Badoff

        Re: Initialisms, NOT acronyms…

        'click'?

  8. Anonymous Coward
    Anonymous Coward

    What about flash foundry capacity?

    So, this seems to be skirting some issues around simple production capacity, and the fact that Flash itself is also reaching some funky density plateaus.

    To make a total flash data center means they need to massive increase the production capacity which is risky from several standpoints. It will drive down prices, which is problematic when you need higher prices to deal with the 6-9 *billion* dollar investment to make a new fab. Without this new production capacity they simply cant make enough to supplant disk for a *very* long time. At which point, who knows what other technologies will be around?

    I don't think either flash or disk is really at risk for a looooong time. My observation has been that storage technologies dont tend to die, the stack just seems to deepen. How long have various groups been claiming tape is dead? Now its just at the bottom of a stack that has multiple layers of disk, flash, cloud back storage (sometimes), etc.....

    1. This post has been deleted by its author

    2. Charles 9 Silver badge

      Re: What about flash foundry capacity?

      "At which point, who knows what other technologies will be around?"

      Thing is, post-Flash at this point is rather like nuclear fusion: it's been "just around the bend" for well over a decade. It's becoming rather a "I'll believe when I see it" thing. When one or two of them hit mass-market release in competitive capacities, THEN we'll pay attention.

  9. Dick Emery

    This will kill X blah blah blah!

    The times we have seen people screaming the death knell of HDD as a storage medium. It's almost become a meme.

    The truth is spinning media has a long history. It's a 'known' medium despite changes to the manufacturing processes to increase density,reliability and speed.

    Flash on the other hand has only been with us a short while in comparison. When you look ahead as a company and look at what it will cost to create new fabs for what is still a relatively expensive medium, with a possibly short period to capitalize on it before something else comes along and wipes it out, I can understand some CEO's hesitance.

    In order for a product to be successful it's not just about all the things we as consumers (including data centers et al) look for in a storage medium. It's about how much return on their initial investment over time they need to make to maximize profits.

    If they plow billions into new Flash and only a couple of years down the line some competitor (notice I did not use 'someone' because as we all know inventors and start ups don't always come to fruition) invents some cheaper alternative that does the same or better. Then they are stuffed (or will at least make a loss).

    For myself I am hopeful of technologies like Memristor. My biggest concerns are:-

    1. Ability to recover data after the drive dies or write cycles have expired (Tests have shown that few if any SSD's allow you to easily recover the data. They should theoretically go into read only mode. But an endurance test done on several brands shows that they pretty much just die and not allow you to access the partitions at all!).

    2. The ability of the device to erase data beyond recovery. Privacy concerns seem to indicate that although it is more difficult to recover data from Flash devices (especially after it has died). The data is actually harder to destroy completely than a simple 'erase' cycle over a hard disk byte value due to over provisioning/garbage collections etc.

    Yes endurance, longevity, speed, reliability and cost are important but without the above two I find myself distrusting the technology for anything other than the operating system and applications. Especially with recent problems like those experienced on the Samsung EVO range (I have a Pro 256GB SDD so it's not so much of a problem although I do notice a speed decrease the more I fill it which I did not realize would be an issue on flash memory).

    1. Charles 9 Silver badge

      Re: This will kill X blah blah blah!

      "1. Ability to recover data after the drive dies or write cycles have expired (Tests have shown that few if any SSD's allow you to easily recover the data. They should theoretically go into read only mode. But an endurance test done on several brands shows that they pretty much just die and not allow you to access the partitions at all!)."

      Because it's not usually the media that breaks on those but the controller. Once that goes, the drive's toast no matter what the tech (even spinning rust is too expensive to reconstruct all but the absolute must-have-to-survive data).

      "2. The ability of the device to erase data beyond recovery. Privacy concerns seem to indicate that although it is more difficult to recover data from Flash devices (especially after it has died). The data is actually harder to destroy completely than a simple 'erase' cycle over a hard disk byte value due to over provisioning/garbage collections etc."

      But at the same time, unless one is willing to physically go and manually inspect every single chip on the thing (a task not unlike electromagnetic restoration of a platter--see above), there's no way to know what's in what. Furthermore, enterprise-grade SSDs can come with internal full-disk encryption to defeat this technique. If the controller goes (or you order a key change), so does the means to recover the data, making the whole "secure erase" business moot.

      1. baka

        Re: This will kill X blah blah blah!

        I don't agree with how hard it is purported to be to erase a drive (controller dependent) For drives with internal encryption, a secure erase just throws away the encryption keys. Without the encryption keys, the data on the drive is useless. Yeah you can see data, but it is encrypted data that cannot be decrypted. Even client class SSDs ship with this today (SandForce is first that comes to mind).

        And yeah, very difficult to recover dead drives. In general if anything in the root area (controller FW / LBA map, etc) becomes corrupted, especially near the end of the life of the drive, then it is toast. In most cases if your primary root record is toast, your secondary (backup) root record is gone too. Its really a matter of the controller croaking rather than the flash as you mentioned. But also when/if a drive enters read-only mode Windows will blue screen because windows likes to write to disks.... and if writes are ignored, windows no likey... either completes boot if this is a secondary drive and doesnt recognize the drive anymore, OR if this is primary, you're duped into believe you can't use the drive anymore.

      2. -tim

        Re: This will kill X blah blah blah!

        I can't pop open an hard drive and read the bit stream using a jtag probe. I can with an SSD.

        It isn't the controllers that fail, its the database the controller keeps about how it mapped the blocks that fails assuming the controller hasn't decided to EOL the drive. If a file system uses lots of extra data to checksum that, the blocks can be recovered and reassembled. There are off the shelf programs that can recover amazing amounts of data from scrambled blocks of even common file systems so I expect that it is easier to recover some data from a broken SSD than a broken spinning disk.

        1. baka

          Re: This will kill X blah blah blah!

          I can use a sata analyzer and record data from host to SSD. big deal. you can read data from jtag, not the same thing. if the controller is encrypting before it is being written to flash, then what you are looking at is meaningless data.

  10. Anonymous Coward
    Anonymous Coward

    Time for another round of Funding then?

    Why else would they send out this [redacted] press release?

  11. Levente Szileszky

    I'm impressed that Violin is still around...

    ...after seeing their uber-pricey yet underwhelming demo 2 years ago then enjoying the reality show-like past couple of years their mgmt produced... how are their revenue streams doing nowadays? :)

  12. yo_G

    bah

    Flash and disk are fine.

    VIOLIN MEMORY IS DEAD.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Biting the hand that feeds IT © 1998–2019