back to article Seagate's lightbulb moment: Make read-write heads operate independently

Seagate is increasing IO performance in disk drives by separating read-write heads into two separate sets which can operate independently and in parallel. The heads are positioned at one end of actuator arms which rotate around a post at their other end to move the heads across the platter surfaces. Thus, with an eight-platter …

  1. A Non e-mouse Silver badge
    Holmes

    Whodathought it - large individual drives suck at random I/O performance.

    I seem to recall this being brought up when 1GB drives were first coming to market: An array of lots of smaller drives has better performance than the same sized array with fewer drives.

    1. Gordan

      '90s Called...

      ... they were please somebody finally caught up with their disk designs. Anyone remember 200MB Conner Chinook dual actuator HDDs?

      https://en.wikipedia.org/wiki/Conner_Peripherals

      1. robidy

        Re: '90s Called...

        Next someome will come up with a technology that doesn't need any moving parts like RAM but remembers stuff when you have a power cut...

      2. Ian Baker

        Re: '90s Called...

        Yes - my immediate reaction was 'This has been done before'

        1. Slap

          Re: '90s Called...

          My immediate reaction was “Failure Rate?”

          1. DougS Silver badge

            Re: '90s Called...

            Failure rate shouldn't be an issue. The previous attempts at this technology replicated the entire set of read-write heads, so you had 2x as many things to go wrong. This simply divides the full set into two, one half can access half the drive, the other half can access the other half of the drive. The same number of read/write heads in total, two actuator mechanisms instead of one but each are smaller - meaning a bonus in terms of faster seek times.

            Basically if you had a 16TB drive with this technology it would perform like two slightly faster 8 TB drives.

      3. gordon123

        Re: '90s Called...

        Beat me to it (and - spookily - with a very similar name).

        Yes, Chinook was the internal name I believe for their high end drive that had two actuators - one at either end of the platter. A bit of a nightmare really as with it being a totally different head stack doing a similar job I think there might have been some tracking challenges.

        Add that to the much higher costs due to 2x head-stacks, and I guess we can see why it never took off.

        I like this idea of splitting the stack though. Not sure why nobody did this before. (Maybe they did..?)

        1. TechnicalBen Silver badge

          Miniaturisation?

          How much of this invention is just waiting for things* to me small enough or reliable enough to make it to consumers? Given enough time, would we not see each arm having it's own actuator? Then I guess it's onto solid state read/write magnet tracks (memresistor)?

          *for others to do the work or doing it themselves. ;)

          1. DougS Silver badge

            Re: Miniaturisation?

            This technology will NEVER come to consumer hard drives, because consumers are no longer the target market for leading edge hard drive development. Consumers are being sold old technology hard drives which will never get the latest technology because the market is too cheap, small and continually shrinking so it cannot justify R&D expenses.

            It wouldn't benefit consumer hard drives anyway, because consumers don't generate enough overlapping I/O 99.99% of the time for this to help. Those few who do generate overlapping I/O should buy an SSD.

            1. Naselus

              Re: Miniaturisation?

              I suspect he may have been using 'consumer' in the sense of 'end user' rather than 'home market'. But yes, there's not really any home user who's ever going to need this kind of performance - this is for NAS devices with hundreds or thousands of concurrent users, who are bottlenecking on drive performance. That's not really likely to occur in the home.

              Still doubt that we'll see every head given an individual actuator eventually, though; it's not that cheap an option and there's a serious diminishing return to the IOPS improvement as e=more heads become independent.

              1. DougS Silver badge

                Re: Miniaturisation?

                Why should there be any diminishing return in IOPS as you increase the number of heads? Due to the smaller size of the actuator meaning it can move slightly faster, you actually give slightly more than n * IOPS for n heads. Throughput increases exactly with n.

                This depends on the drive having enough I/O requests coming in to keep it busy, but that shouldn't be a problem for cloud providers where this is targeted at. It would make much less difference for consumer drives in a PC (which it will never be offered to) because their drives at idle almost all the time.

                The other advantage is that RAID rebuilds take place more quickly - n times faster based on n heads (exactly n since it is a sequential process)

                1. Naselus

                  Re: Miniaturisation?

                  "Why should there be any diminishing return in IOPS as you increase the number of heads?"

                  There's not a diminishing return in the IOPS (provided you're always running the disk at maximum). There's a diminishing return in the relative improvement you're seeing over the previous upgrade.

                  So if you have the traditional 1 fixed set, you have performance X, and spliutting the stack in half basically doubles your output to 2X ( plus a fairly negligible increase from weight reduction). That's a 100% performance improvement for the price of the additional actuators. But adding a third arm only takes you from 2X to 3X, so that's only a 50% improvement over the 2X.

      4. Peter 39

        Re: '90s Called...

        The 90's ... really ???

        No. This is a 1960's scheme, with giant platters and one head for each surface.

      5. RLWatkins

        Re: '90s Called...

        The '90s? Remember those washing machine-sized 20MB top-loading unsealed multi-platter drives that old DEC and DG boxes used? I recall some DEC guys experimenting with putting four sets of actuators and heads in those in the mid 1970s.

        Cool idea, and interesting trying to come up with software which would take a read/write queue and optimally route the requests to the appropriate heads... on a machine with a 64K address space, no less.

        This profession has always been a blast, and stays that way - as much fun as you can have with your pants on.

        1. Anonymous Coward
          Anonymous Coward

          Re: '90s Called...

          This profession has always been a blast, and stays that way - as much fun as you can have with your pants on.

          You still wear pants? That is why we telecommute....

    2. Blank Reg

      Not only performance improves, but rebuild time when one (or more) of your 10+GB drives fails in a RAID array are horrendous. Though power consumption and required space will be higher with all those tiny little drives.

  2. frank ly Silver badge

    Worth the effort?

    They seem to have packaged two physical drives in one physical unit. Maybe the on-board controller can also treat them both as one physical drive but isn't it more flexible and perhaps cheaper to concentrate on making a single physical hard drive robust and cheap and then use RAID and similar techniques to get performance increases?

    If you want to go this route, why not package an extra entire independent set of read/write heads at the opposite corner, then you could really have two channel access to the same physical drive?

    1. Anon

      Re: Worth the effort?

      I suggested the sets of heads on opposite corners a long time ago. Double the rate of read-after-write verification. Read an entire track in half the time. Zero adjacent track-to-track seek time. But would they listen? Noooo.

      1. Geoff Campbell

        Re: Worth the effort?

        Someone (Conner, perhaps?) did try such a thing. It didn't catch on.

        Realistically, we're in the end game for spinning rust now, so big new innovations seem a little pointless to me.

        GJC

        1. Geoff Campbell

          Re: Worth the effort?

          Ah, yes, here we are:

          http://www.tomshardware.co.uk/seagate-hdd-harddrive,news-31531.html

          GJC

          1. Peter2 Silver badge

            Re: Worth the effort?

            "Realistically, we're in the end game for spinning rust now, so big new innovations seem a little pointless to me."

            Mmm.

            Personally, my storage is m2 SSD, PCIE SSD and a couple of large drives for keeping large amounts of data. I agree that SSD's are the future.

            However, i also think that given the production costs and capacity of memory for SSD's we will still have HDD's with us in 20 years time. Why?

            If your Dell/HP/Fijitsu/etc making a budget desktop, then what's it going to take to replace the HDD? A 500GB HDD can be had for £25 at retail prices. Large OEM's will get it cheaper than that. An SSD is over quadruple this price. The SSD will kill the HDD when the price reaches parity.

            That won't happen absent a new fabbing technology for memory chips that reduces the cost by around 400%. Ergo HDD's are going to hang around at the low end of the market for cheap drives, and for large scale storage where it can win in cost per GB over SSD.

            The "end game" for HDD's is sort of like the "end game" for tape. Yeah, it's going to get pushed into niche markets but it's going to survive there for decades, and some of those niche applications (like enterprise storage on SANS) are massively huge.

            1. Naselus

              Re: Worth the effort?

              "Yeah, it's going to get pushed into niche markets but it's going to survive there for decades, and some of those niche applications (like enterprise storage on SANS) are massively huge."

              Agreed. If you look beyond end-user storage (which most commentards apparently struggle to do) and look instead at worldwide storage requirements, it's clear that Flash remains a minority and is not expanding as quickly as data storage is. Flash may dominate your options at PC world, but it's not what big organizations are storing their petabytes of data on.

              I suspect that, long-term, NAND might be in more danger than spinning rust, simply because people are actively looking for something faster than flash and it's very expensive to produce (and even more expensive to produce the fabs to produce it). If something out-performs it, Flash suddenly becomes second-rate tech with a heavy price tag, and no new Flash foundries will be springing up to push that price down. On the other hand, the only reason I'd replace most of my slow-tier 7.2k disk is if something much cheaper appeared with the same performance - rather than something much more performant at the same price. No-one is really throwing much effort into finding that, since disk is already cheap.

              Flash has probably already eaten all the disk market that it's going to, basically wiping out the 15krpm and 10krpm classes completely. But the things Flash does well are not the things 7.2k rpm does well, and as a storage admin there's a lot of situations where I wouldn't even consider Flash as an option.

    2. Warm Braw Silver badge

      Re: Worth the effort?

      Or just go the whole hog?

  3. Doctor Syntax Silver badge

    Wouldn't an 8 platter drive need 16 heads or are they just using single sides these days?

    In effect this splits the drive into two logical drives. There needs to be a prize for someone who has the bright idea of using these as a mirror pair. A wooden spoon?

    1. AstroNutter

      Could use them as a stripe set.

      But then why didn't that do that already in the drive controller? Not like there's going to be any other advantages, other than two drives in the space of one. You still get the problem of one drive fails, both need replacing.

    2. John Robson Silver badge

      I need another wooden spoon for the kitchen...

      And this was my first thought.

      Why involve any change externally. With the size of drive caches nowadays the disk is doing all the data rearranging anyway, so it could reasonably intelligently divvy the data up itself.

      Of course a naive striped pair approach would simply tie the heads back together again (on write at least), but an internal firmware change to take advantage of what I suppose is effectively 'more spindles' strikes me as an easier approach than most other options (unless you genuinely present as two disks (can a normal SATA/SAS port take that?)

  4. Anonymous Coward
    Anonymous Coward

    What's the point?

    I can maybe see the point if you're a hyperscale cloud provider, but surely in this day and age if you need more IOPS on premises you're just using flash? I've even got plenty of customers using flash for bulk serial IO now as well because it's cheaper to run one type of disk than two.

    I just don't see where a disk that produces maybe 1.5-2x more IOPS for the same bandwidth fits into a modern architecture.

    1. Anonymous Coward
      Anonymous Coward

      Re: What's the point?

      > I just don't see where a disk that produces maybe 1.5-2x more IOPS for the same bandwidth fits into a modern architecture.

      It could be double the IOPS *and* double the bandwidth, as long as there are two sets of electronics which can read/write to both halves simultaneously. People who don't need the IOPS could use it as a RAID0.

      Having said that: I never understood why hard drives don't "stripe" across all heads. If you have 4K sector size (which is common in modern drives), and 8 heads, you could write a 512 byte chunk on each head simultaneously. You waste some capacity in inter-block gaps (which is what the 4K sector size was supposed to fix in the first place), but you get 8 times the throughput.

      1. really_adf

        Re: What's the point?

        Having said that: I never understood why hard drives don't "stripe" across all heads.

        AIUI head alignment requires constant adjustment so you can only have one head aligned at a time if they are mechanically linked. Unless, I guess, you massively reduce density and hence capacity.

        1. Dwarf Silver badge

          Re: What's the point?

          Having said that: I never understood why hard drives don't "stripe" across all heads.

          They do, its called a cylinder, you look down the heads across all surfaces and write across them,

          See CHS Hard Disk Addressing

          No reason to believe that the vendors stopped doing this when they moved the external interface to LBA back around 1980 ish

          Unlike what the other post on this thread claims, its not hard to analyse errors on a per surface perspective, since the drive's controller and firmware knows the real geometry of the drive and grown defects do just that, so you can easily determine if the failed sector is adjacent to an existing defect.

          None of this is new technology.

      2. DJO Silver badge

        Re: What's the point?

        Having said that: I never understood why hard drives don't "stripe" across all heads.

        Hard drive are not as infallible as you might think. They all have a overhead of capacity which can be brought into use if any sectors start to look a bit iffy, this is long before the SMART stuff kick in.

        If you striped across all platters a duff sector on one platter would effectively mean a duff sector on all platters which would eat the capacity too quickly.

      3. inmypjs Silver badge

        Re: What's the point?

        "why hard drives don't "stripe" across all heads"

        Because a head is servo positioned on the track it is reading or writing and the alignment between head and track on other platters is no where near good enough. For example at current densities a 1 degree C temperature differential between two aluminium disks would produce a thermal expansion differential about 6 tracks wide at 1 inch from the spindle.

        I would speculate micro (probably piezo) actuators on each arm could accommodate errors and allow one head on each arm to servo onto the 'same' track. Would need a lot more electronics to achieve.

  5. Alan Brown Silver badge

    about the only possible use that I can think of

    Is to have them seeking in equal but opposite directions so the torque moments cancel.

    Being able to read/write simultaneously on all heads would be more of an improvement (they're currently accessed sequentially)

    1. DougS Silver badge

      Re: about the only possible use that I can think of

      They can't read/write on multiple heads at once because the tracks are so close together that thermal variation means that all heads don't align with the same tracks every time. They'd have to reduce density by a fairly significant margin to do what you suggest.

      1. Alan Brown Silver badge

        Re: about the only possible use that I can think of

        I'm aware of why they don't work in parallel, nonetheless it would still be an improvement if they could.

        The next logical step from what this proposition is to move to an independent arm per platter (2 heads).

        Personally I don't think this will work well enough to leave the lab.

        Apart from the issue of conflicting magnetic fields on the other side of the pivot (ie the field driving one arm will perturb the other one), I suspect that turbulence issues will raise their ugly heads in short order. There's a lot of funky hydrodynamics at work keeping a hard drive's heads afloat and they tend to assume smooth airflow.

    2. DougS Silver badge

      Re: about the only possible use that I can think of

      Doubling bandwidth and IOPS isn't good enough, you think the only use is cancelling torque moments? Each mechanism would weigh only slightly more than half the weight of the full (current) mechanism, so even when they both swing the same way the torque is barely greater than it is now.

      What I wonder is whether the movement of the second set of heads will cause vibration that throws off the first set when it is actively reading and writing. It may require more vibration damping than current drives, but that's fine since this is not targeted at moving devices like laptops, and they can easily mandate specific mounting requirements that would be infeasible for consumer targeted PC drives.

  6. RockBurner

    isn't the logical conclusion to this for each individual disk to have 2 independant arms (one read, one write).

    Obviously you'd have 2 pivot posts (not impossible), with each arm operating independantly on the post (I suppose it depends exactly where the electric motor actuating the arm movement is located).

    the main challenge would be packaging it all up, rather than anything else.

    1. Aladdin Sane Silver badge

      The fluid dynamics would be interesting. Currently, heads 'fly' on a cushion of air generated by the spinning disk. If 2 pivot posts are in use, the heads might interfere with airflow and insufficient lift may cause head crash.

    2. DougS Silver badge

      No, separate arms for read and write would be far inferior to what they've done. This allows them to double read and write bandwidth (assuming you have enough I/O that each 'half' of the drive is fully utilized) and more than double IOPS. Yes, more than double, because you not only have two sets of heads, each accessing half the drive, but since each mechanism has half the heads it weighs less which means seek times are slightly faster.

  7. Prst. V.Jeltz Silver badge

    This is why all cars have 2 camshafts these days. Lightbulb moment.

    1. Unicornpiss Silver badge

      Disagree..

      "This is why all cars have 2 camshafts these days. Lightbulb moment."

      Actually not nearly all cars have or need 2 camshafts. Including some quite powerful ones.

  8. Anonymous Coward
    Anonymous Coward

    Well that's novel. Or maybe not.

    See e.g. Conner Peripherals product codenamed "Chinook" (it's on Wikipedia).

    And long before Chinook there were things that can probably be found in places like Jim Austin's Computer Collection; standard (for the era) size disk drives that would fill a 19" rack or more, with multiple independently operatable disk heads. My memory is failing, I cannot quickly find any particular examples of such a product, but they did exist.

    1. RealBigAl

      Re: Well that's novel. Or maybe not.

      Have an upvote for Connor Peripherals. I remember them in some old Compaq machines.

    2. Alan Brown Silver badge

      Re: Well that's novel. Or maybe not.

      " with multiple independently operatable disk heads."

      There's a reason you don't see them anymore - and can't find museum examples. They were unreliable.

  9. ThaumaTechnician

    Dang!

    Here I was, looking at the image thinking: "Wow! They're using a set of write-only platters and another set of read-only platters? I'd like to see a graph of performance vs. cost..."

    /really, this is worth a headline?

    1. Norman Nescio Silver badge

      Re: Dang!

      Perhaps I'm missing something here, but wouldn't that graph would be a flat-line at zero? My thinking being, "How would the data get off the write-only platters, and data onto the read-only platters?".

      On reading the headline, I thought they had optimised one set of heads for reading and one set for writing and given them separate actuators, which couls have been interesting.

      1. ThaumaTechnician

        Re: Dang!

        How? The Signetics 25120 to the rescue.

  10. Alister Silver badge

    Make read/ write heads operate independently (on seperate platters)

    So, not actually, then.

  11. Daniel Hall
    Facepalm

    I cant be the only one..

    ..that initially read the heading and thought, oh cool, one actuator for reading and another actuator positioned somewhere just for writing...

    No?

    1. Doctor Syntax Silver badge

      Re: I cant be the only one..

      Followed rapidly by "Wouldn't it be better to have them both read/write?" heads. And that followed by "That's what they've done". Followed in turn by "But they've only addressed half a cylinder at a time. Why didn't they make two separate arm sets and make them full height?".

      1. DougS Silver badge

        Re: I cant be the only one..

        If they were full height they'd weigh more, cost more, and wouldn't get the advantage of half height actuators - better seek times to take the doubled IOPS and more than double it.

        This strategy could only work for drives expected to be busy, as you'd see for a cloud provider or other markets where you have a lot of data but don't need SSD-level instantaneous access times. If the drive is often maxing out its I/O capability this improvement will double the MB/s throughput and more than double the IOPS.

        Theoretically they could split them again for another doubling of throughput and more than doubling of IOPS, though there would be diminishing returns at some point from all the separate motors meaning increased cost and reduced MTBF.

  12. Martin Gregorie Silver badge

    Old tech solved this decades ago

    ICL solved this problem back in the early '70s in software for the George 3 OS from version 6 onwards. No hardware changes needed.

    The solution used two i/o request queues for each disk plus a record of where the heads were and which direction they were moving (inward or outward). Requests were added to a queue in positioning order. Which queue they were added to depended on whether they could be serviced by continuing to move the head in the same direction or not.

    The end result was that head movements were much shorter than if they were serviced as they arrived and consequently overall throughput rose. The improvement was obvious if you stood by a working drive: instead of the heads thrashing in and out, they floated relatively slowly in and out across the disk accompanied by a steady, and much faster, stream of clicks for each head movement. IIRC this at least doubled the per-disk throughput.

    Of course, this approach won't work if there is only a single task running fast enough to dominate disk i/o, but for any server with a multiprocessing workload it should help a lot.

    1. Doctor Syntax Silver badge

      Re: Old tech solved this decades ago

      Isn't that what disk drives handle themselves these days? https://en.wikipedia.org/wiki/Elevator_algorithm

    2. NeedBongos

      Re: Old tech solved this decades ago

      IBM 3380 had 2 read/write actuators per drive, needed them to make S/38s go.

      First UK deliveries were in 1982.

      https://en.wikipedia.org/wiki/History_of_IBM_magnetic_disk_drives#IBM_3380

    3. Anonymous Coward
      Anonymous Coward

      Re: Old tech solved this decades ago

      "a record of where the heads were and which direction they were moving (inward or outward). Requests were added to a queue in positioning order. "

      Isn't that a version of the "elevator seek" algorithm (or just "elevator algorithm")?

      http://www.cs.iit.edu/~cs561/cs450/disksched/disksched.html

      [edit: apologies to Doctor Syntax, you said this before I did]

    4. Frederic Bloggs

      Re: Old tech solved this decades ago

      Perhaps you were thinking of ICL CAFS disks. Essentially the drives had 16 heads per arm instead of just the one. The algorithms in the controllers were also designed to aid the searching of content addressable data, but it was the fact that there were 16 channels of streaming data available that made the thing blindingly fast.

      1. DougS Silver badge

        Re: Old tech solved this decades ago

        You can't have more than one head per arm active in the drive (not just on a platter, in the entire drive) because the tracks are far too close together. Thermal variation means that you can only line up properly with one track on one platter/surface at a time.

    5. Alan Brown Silver badge

      Re: Old tech solved this decades ago

      "The solution used two i/o request queues for each disk plus a record of where the heads were and which direction they were moving "

      ZFS achieves something similar by storing up writes and periodically despooling them sequentially. It does as much read reordering as it can and the ARC + L2ARC function aims to keep commonly used data/metadata where disk seeking isn't needed.

      Every time someone's tried mechanical cleverness in hard drives it hasn't ended well. Simpler mechanics and more complex software is almost always a better fit.

    6. bombastic bob Silver badge
      Devil

      Re: Old tech solved this decades ago

      "The solution used two i/o request queues for each disk plus a record of where the heads were and which direction they were moving"

      for this to have ANY kind of performance advantage, you'd need an OS that's aware of this feature, and could internally prioritize things for best throughput. But for desktops, at least, I'm pretty sure that Windows operating systems force a physical write to the disk way too often, and seem to be more like DOS in their "serialized" way of doing things than like a REAL operating system (say Linux or BSD, k-thx).

      maybe a massively parallel or massively "async I/O" process would benefit, sure, but not your average desktop application. Users wouldn't see any difference, in other words, to justify the price.

      it's like multi-core in a way - very few, if any, desktop applications even REMOTELY come close to leveraging this. And yet nearly every CPU sold nowadays is dual core or better. And is the *PERCEIVED* speed of those "modern" OSs (say Win-10-nic) any better than, say, XP or 7? *NOT* *BY* *MY* *MEASUREMENTS* !!! [and Micro-shaft's "OS background schtuff" is at LEAST a little disturbing anyway]

      /me wonders just how much ZFS knows about the way the data is stored on the drive...

      1. DougS Silver badge

        Re: Old tech solved this decades ago

        This isn't designed for desktop PCs, and may not even work for them - I suspect the tolerance for vibration would be way less for these drives so they might have specific mounting requirements. You probably won't want to slap them in a case with metal screws on a metal bracket and absolutely wouldn't want to stick one in a laptop.

        I don't think the OS needs to know anything about this, it would be presented to the OS as two LUNs in a single drive as if it is a small RAID array. There isn't really any need for OS support, since there's zero overlap on the drive between the two arms. If there are any specific restrictions from interference between the two arms moving at once, or one moving while another is writing etc. it would have to be handled in the drive's controller anyway. The OS won't need to know anything special to get optimal performance from these drives. All it will need to know is that both LUNs are really on the same physical drive, so it won't let you do something stupid like trying to mirror them together.

  13. Mage Silver badge

    Half the drive's recording heads will operate together

    I'd imagined two sets, but I suppose it would make the drive longer.

    1. Alan Sharkey

      Re: Half the drive's recording heads will operate together

      Why stop at 2 sets? Have 2 heads per platter, one for read, one for write.

      1. Anonymous Coward
        Anonymous Coward

        Re: Half the drive's recording heads will operate together

        Why limit each head to a read or write, if the heads could do both, it would double the read and write performance on read or write intensive tasks.

  14. Lotaresco

    Rotational latency

    Now, if only someone could come up with a technology that has one read/write head per track or preferably multiple read/write heads per track thus eliminating positional latency and reducing rotational latency. Oh they did, in 1932.

  15. Anonymous Coward
    Anonymous Coward

    Reliability

    I'd rather see Seacrate invest some R&D in making their drives a bit more reliable. They're cheaper than other makes for a reason.

  16. Anonymous Coward
    IT Angle

    Teahound's lightbulb moment...

    Store less shit.

  17. foxyshadis

    Silly suggestion

    Why on earth would you need to expose it as two drives? SATA/SAS already queue up tons of requests and the drive is already allowed to service them in non-linear order, as long as it's within the timeout. That's one of the pivotal parts of AHCI that makes it a huge improvement over ATA (Legacy) mode. If you have a parallel workload that wouldn't benefit from the improved random workload, then you can gain no benefit out of the dual heads at all anyway.

    Displaying it as two hardware drives just sounds like a good way to confuse the hell out of most operating systems. Just internally split it into zones of some megabytes each, that'll nicely split up data. I suppose include an initialization command so that the OS can see both if it REALLY wants to micromanage it.

    1. DougS Silver badge

      Re: Silly suggestion

      They wouldn't split it as two drives, they'd present it as two logical volumes on the same drive. Software already has support for that - it would look like a RAID controller with two volumes.

      1. foxyshadis

        Re: Silly suggestion

        Two volumes is what I mean by two (logical) drives, it's exactly the same scenario: It pushes the logic all the way up to the application or OS, which still won't be any good at handling it without specialized knowledge of the drive it's interacting with -- when was the last time you saw an OS or application that was any good at scattering files across multiple volumes evenly? Most of them will just store all the most-accessed stuff on one and hardly anything on the other, reducing access times instead of raising them.

        Whereas they could just stripe every couple of megabytes and create a reasonable default, and if they really wanted to go hog-wild, keep statistics to try to even out access patterns over time by moving files around disks.

        1. DougS Silver badge

          Re: Silly suggestion

          Why should the OS need to know anything? It should just stripe them together as you say, so that big files can be read/written simultaneously from both halves, and small files can be accessed independently on each. It just needs a way to know that both LUNs are really part of the same physical drive so it can stop people trying to mirror them, kind of like how an OS needs to know that two threads are part of the same physical CPU core for scheduling reasons.

          It is no different than how you get the best performance from two separate drives - you stripe them together. Except in that case you increase the chance of failure since they are two separate drives whereas in this case it is all one drive so you don't even take the reliability hit.

          Now that I think about it, it might just stripe them together internally and present it as one big drive, at least as a default, since that's the only reasonable thing you can do with it anyway. The only reason you'd want them separate is so you can fiddle with the stripe size if you know something about the access patterns and typical I/O sizes.

  18. Anonymous South African Coward Silver badge

    Seagate IronOctopus, soon to be seen at a reseller near you.

  19. networkboy

    Or the of 160MB WREN iii ESDI drives. Those guys had linear voice coil actuators on both sides of the drive. Showed up as two LUNs on the ESDI controller.

  20. Jim O'Reilly
    Holmes

    This has been done before. Hitachi tried it in the 1980's. Not much interest!

    Anyway, this improves IOPS from 150 per secomd to maybe 300 per second...only 44,700 more to catch up with an SSD!

  21. Androgynous Cow Herd

    I bet the guy who built the very last buggy whip...

    I bet the guy who built the very last buggy whip built a really compelling, state of the art buggy whip. But the market had moved on to automobiles.

    Improving I/O on spinning media these days is a bit silly. Incremental gains here are not going to move the market back from he newer solid state technologies to spinning media.

    Spinning media has it's place, for workloads where capacity is more important than latency. Otherwise, Flash media continues to spiral down in price and will run rings around this from an I/O latency standpoint.

    Neat idea, though.

    1. DougS Silver badge

      Re: I bet the guy who built the very last buggy whip...

      Hard drives have at least a decade longer to run, because SSDs are a long way from beating them on $/TB. They no longer make sense anywhere you care about performance, but there is a lot of bulk/cold data storage, or backup to disk, where paying extra for SSDs makes no sense.

      If you had 10 or 20 TB of 4K video of your kids you recorded in the next couple years, would you really pay thousands of dollars for SSDs to store them on when you know you'll never look at most of the files ever again, and those you do will be rarely accessed and not benefit from shaving 10 ms off seek time and getting 600 MB/sec instead of 100 MB/sec? Besides, SSDs won't hold their data for years sitting on a shelf unpowered, but hard drives will.

  22. razorfishsl

    That's GONNA get HOT.

    Two voice coil magnets for moving the heads....

  23. Lamb0
    IT Angle

    One of the problems with the Conner's dual servo drive was vibration. The opposing arrays each covered all of the platters' sides. The Bernoulli Effect isn't powerful, yet, it's powerful enough to induce vibration on the platter. With multiple heads on the same platter sooner or later a resonance can be induced significant enough to crash of opposing heads. Long term reliability was NOT good.

    Seagate uses separate servos and head arrays for different platters to sidestep much of the unbalanced asynchronous force inducing vibration and may be much more reliable. The dual servo action rather reminds me of scissors, which might also reduce unpredictable vibrations. (If they can be predicted, they can be compensated.)

    I see no reason for RAID1 - I'd prefer a separate drive for mirroring. While an enterprise might want the dual servo arrangement for more I/Os, that's where Flash rules. For archival purposes, however, sustained SATA3 speeds from near SATA2 technology can speed backups and halve rebuild times for (hopefully) a modest increase in the price of the drives. Would you rather have a ~12 to 16TB drive read in nearly a day, or maybe only a shift?

  24. Anonymous Coward
    Anonymous Coward

    Hardly a lightbulb moment, doubling the number of heads, doubles the cost. This is purely an attempt to remain relevant in the days of SSDs

  25. Lamb0
    Facepalm

    I still don't see it as a valid attempt to compete with SSDs using Ye Olde spinning rust. In case you haven't noticed, the head count matches the number of platter sides. Dual actuators with only one set of heads per platter should NOT double the cost, but SHOULD at least double the throughput from a similarly sized drive. WIN/WIN

  26. Sssss

    What! Multiple heads in a drive. I thought of that decades back. Why aren't I a billionaire.

    Lol. :) You should have seen my finale solution. I wanted to use drives as RAM.

  27. Anonymous Coward
    Anonymous Coward

    RAID 5?

    How about RAID 5 in a single hard drive. Three sets of read/write heads...

    1. sidusnare

      Re: RAID 5?

      You do know what the I in RAID is for?

      1. Anonymous Coward
        Anonymous Coward

        Re: "You do know what the I in RAID is for?"

        "You do know what the I in RAID is for?"

        Yeah. Inexpensive.

        Your point?

        Please don't suggest that I shouldn't be mirroring data across two platters/volumes/etc in the same drive, it's never faillkdjf;klv xcv df;lkasrgp[y

        ;)

  28. sidusnare

    Spinning rust

    Those dismissing this in favor of SSDs, until SSDs surpass spinning disk in terms of price per The at the same density, people/ companies that work with any sizeable amount of data will always want more of this.

    1. Sssss

      Re: Spinning rust

      Also, you can crack open drive and repair it and do data recovery.

  29. sloshnmosh

    dd if=/dev/zero

    Although I am a big fan of how fast my SSD's are, I did like the fact that I could overwrite free space and verify it with sed on a platter type drive.

    Is there a way to force garbage collection on a SSD?

  30. Anonymous Coward
    Anonymous Coward

    Perhaps

    Installing two full sets of heads 180 degrees apart shouldn't be too hard to implement, or 4 sets at 90 degrees. Software would do the rest.

  31. Dave 13

    Zaphod

    Friends don't let friends buy spinning rust - no matter how many heads.

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