back to article Cool disk drive actuator pillar, Seagate – how about two of them?

As we've previously written about here, Seagate has the technology to split a disk drive's read-write heads into two independent sets, an upper and lower half, which increases the drive’s IO performance by having the upper and lower platter sets do IO in parallel. Disk drives with two actuator posts could read and write in …

  1. Anonymous Coward
    Anonymous Coward

    IIRC some disk drives in the 2nd generation mainframes had two sets of heads. One set was fixed for the outer track(s) and reserved to give system software fast access.

    1. Warm Braw

      Drives that were intended for paging sometimes had multiple sets of fixed heads and were frequently referred to as "drums" by analogy with their historical counterparts despite being platters rather than cylinders.

    2. Gordan

      TL;DR

      It won't make any difference to sequential I/O, including RAID rebuilds.

      It will, however, double the throughput on random I/O.

      This applies to both reads and writes, all that matters is whether the I/O is sequential or random.

      1. Anonymous Coward
        Anonymous Coward

        Re: TL;DR

        I will also, however, lead to some idiot putting the "mirror" of his RAID 1 on the second logical drive, with predictable outcomes.

    3. Stoneshop

      Two sets of heads

      DEC's RA8x series drives used two heads per side on a single arm, effectively halving the seek stroke while still being able to access the entire platter.

      I've also encountered a drive with two independent positioner assemblies, but details appear to be in some offline part of my memory.

      1. Anonymous Coward
        Anonymous Coward

        Re: Two sets of heads

        Conner Peripherals Chinook, two head assemblies, each on its own actuator.

    4. vincent himpe

      The fixed head was used to read the sync track. the sync track holds the markers where sectors begin and end.

      Early drives did not use soft sectoring but had hard sectors.

  2. Anonymous Coward
    Anonymous Coward

    Two sets of heads moving independently and concurrently will give some interesting vibration interactions.

    If they are top/bottom of one post then it will be "bent" in interesting ways (albeit by very tiny amounts).

  3. Joerg

    What is Chris Evans babbling about there??

    WHAT ??? "“How will data be written to the drive? If we want to double throughput, both sets of platters would need to be written at the same time. This would negate any performance improvement on read, because all heads would be dedicated to reading that data and couldn’t do other work.”"

    Was Chris Evans high or what? What he said there doesn't make any damn sense. It replied like someone with no basic knowledge of RAID ... Obviously the data will have to be spread on platters like it was two RAID-0 units. Also which Hard Disk drive has to date been capable of doing anything else at the same bandwidth with no massive slowdown when reading or writing ? Try copying or moving a 20GB+ file (or many small ones) from an hard disk to another and at the same time reading or writing other files on the same disk... it would just slow down as hell ! Only with expensive RAID controllers it is possible to do that mainly due to caching.

    1. Anonymous Coward
      Anonymous Coward

      Re: What is Chris Evans babbling about there??

      My issue is his claim "I’m not sure that this is right, because if you have (say) 8 heads reading/writing compared to 2 groups of 4, how is the throughput any higher with purely sequential?"

      Clearly he doesn't even know how hard drives work, because you NEVER have more than one head reading or writing at once in current hard drives. Yes, there is one head for each surface (i.e. two per platter) but only one can read/write at once. There's no way to have more than one do it because the tracks are so narrow that it is impossible to line up more than one head at a time, due to uneven heat expansion. If you have two sets of heads that each operate over half the platters then you will have TWO heads reading or writing at once, which will double throughput and double IOPS (actually very slightly more than double IOPS because each mechanism has a bit less mass and thus should be able to seek a bit faster)

      How does someone get the title "Storage Architect" when they don't understand the basics of how storage devices work? Someone sent him to enough EMC classes and he got a certificate, I guess.

  4. Anonymous Coward
    Anonymous Coward

    One read write head per track ?!

    What about one read/write head per track ?

    i.e. The read/write arm doesn't move atall, it just sits across the radius of the drive with a read/write sensor above each and every track on the disk.

    (C) Bahboh 2018. Patent pending.

    1. Anonymous South African Coward Bronze badge

      Re: One read write head per track ?!

      You, sir, are a genius!

      Oh wait...

      How many tracks does the average hard drive have? Let's assume a 3.5" 2Tb HDD. Can anybody work out the number of tracks on a single platter?

      1. Anonymous Coward
        Anonymous Coward

        Re: One read write head per track ?!

        Why do you think the number of tracks matters ?

        The read write head can surely be as narrow as the track it sits above ? (even assuming much smaller track widths since 2006 )

        "A typical magnetic region on a hard-disk platter (as of 2006) is about 200–250 nanometers wide (in the radial direction of the platter)"

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

        1. Pascal Monett Silver badge

          @Bahboh

          From the article you linked, there is a reference to the heads that indicates that read/write heads are 0.3 mm wide.

          That's already damn small.

          Logically speaking, you might be able to get the read/write portion of the head down to a tenth of that size, but you still have the arm that carries it - and that thing will remain a lot bigger because of the physical constraints.

          1. vincent himpe

            Re: @Bahboh

            The pole tips are much smaller than that. In the micron order. The pole tips compress the magnetic field so it runs into the track. there is substantial leakage into adjacent tracks, but due to the magnet hardness the leakage is not strong enough to 'flip' the data in the adjacent track.

            Even if you could make such an array for reading , for writing it is not possible. the currents involved and the size of the coils ( writing is still done using a coil , reading is done using a magnetoresistive element ) make it impossible. Try making a transmitter that can blast 4 to 6 GHz at 400 to 600 mA of current....

            on a few million outputs at the same time... that chip would get so hot it would melt down ...

            note : i have designed harddisk read/write preamps as well as motor controllers...

            1. Anonymous Coward
              Anonymous Coward

              Re: @Bahboh

              If it could be done for only the read heads with a seperate, moving, write head, that would be useful, no ?

              Also, one could place, say, 8 arms around the disk, each with fixed read heads above every 8th track so the heads could be 8 times wider than a track ?

        2. Stoneshop
          FAIL

          Re: One read write head per track ?!

          Why do you think the number of tracks matters ?

          Because the actual head is about 1000 times as wide as a single track, meaning there's NO FSCKING WAY you can get separate heads over each track.

          Plus there's the cost.

          With your first comment I thought you were trolling, but it appears you really think it's feasible. It is not.

    2. Anonymous Coward
      Anonymous Coward

      Re: One read write head per track ?!

      It might have to do with the cost of that head. Given the engineering involved, that head's probably the most expensive part of the armature. Now multiply that by a few hundred.

  5. David Knapman

    For two independent sets of heads to be able to work with the same sets of tracks, the positioning needs to be spot on. When you have one set of heads, it doesn't matter *precisely* where the heads go when they're looking for track 13, so long as they consistently move to the same physical position each time track 13 is asked of. But with two sets of heads, they also have to agree on what that position should be.

    I remember that, years ago, this was the reason given for the much lower densities on removable media than were available with hard drives - because they were, effectively, subject to being interacted with by many sets of heads over their lifetime and so the tracks needed to be wide to allow for misalignment.

    Perhaps this is less of an issue these days - not sure.

    1. ilmari

      I thought harddrives were more like optical drives these days, that they actually track the position of a track, so that it allows for the disc/platter to even spin slightly off-centre.

    2. Stoneshop

      When you have one set of heads, it doesn't matter *precisely* where the heads go when they're looking for track 13,

      That would be a sticking point if the positioner controller was using an external reference, like servo tracks or even a feedback system coupled to the positioner itself, but that has long passed. Nowadays the positioning data comes from the active head itself, in a way that you might want to describe as 'maximum correct signal' (not unlike manually tuning a radio). Although the drive logic has to have some rough idea of where track 13 is, and how much to seek inwards or outwards to get there, the final positioning over the track is done using the signal from the head itself.

      I remember that, years ago, this was the reason given for the much lower densities on removable media than were available with hard drives

      That's part of it. Another reason is that the environment in a drive using removable packs can't be as clean as in a HDA, so head flight height can't be as low. And track width is related to head flight height again: lower height means narrower tracks means greater data density.

    3. inmypjs Silver badge

      "consistently move to the same physical position"

      Heads haven't moved to the same physical position for decades.

      They find a signal from a track to servo align themselves to then read data from the track to determine where they are. Knowing where they are they can guess at what is needed to jump to the next place they want to be.

  6. Anonymous Coward
    Anonymous Coward

    I'm Praying to the Aliens..

    Maybe spinning metal discs are the last word in huge archival storage for now (perhaps not) but jeez, isn't it time these fragile, archaic things were consigned to history.. No, i have no idea what to replace it with, having to immediately access quillions of bits of data immediately is no mean feat, i rather think that dealing with such huge volumes of data at such a minute level really seems to be heading up a blind alley somewhat, and the whole way the architecture deals with data needs a rethink..

    1. Pascal Monett Silver badge
      Trollface

      Re: I'm Praying to the Aliens..

      Don't worry. Holographic storage will be here in 5 - 10 years.

      Just keep repeating that to yourself and all will be well.

      1. Korev Silver badge
        Joke

        Re: I'm Praying to the Aliens..

        Yes Sir - performs a "Rimmer salute"

    2. phuzz Silver badge
      Coat

      Re: I'm Praying to the Aliens..

      Data will last for ages on sheepskin parchment, all you need then is very neat handwriting and lots of patience.

      >>> mines the one with a quill in the pocket.

  7. Adrian Harvey
    Boffin

    Does HAMR / MAMR impose any limitations?

    Thinking on this concept - in my crude mental model of heat assisted recording - where heat is used to reduce the coercivity of the magnetic surface material in a very narrow track so that the written track can be narrower than the write field from the magnetic part of the head - the disk will need some portion of the rotation time to cool down. If, having two sets of heads, writing neighbouring tracks, there may be some unwanted crosstalk if the track being written by the first heads hasn't cooled fully by the time It reaches the second heads writing an adjacent track.

    Does that make sense? And how much of a problem would it be really? I'm not sure how much cooling is really needed... or if MAMR would offer any advantage here...

  8. ilmari

    So...

    So.. Am I missing something, or does Seagates's split actuator thing perform the same as two half-sized drives in striped raid-0?

    1. Stoneshop

      Re: So...

      Theoretically yes, the way I read it. At least, as long as you're not (close to) saturating the host interface and the drive buffers.

    2. Doctor Syntax Silver badge

      Re: So...

      "Seagates's split actuator thing perform the same as two half-sized drives in striped raid-0?"

      Yes, providing you don't mind both of them failing at the same time.

      1. Stoneshop

        Re: So...

        Yes, providing you don't mind both of them failing at the same time.

        With RAID-0, they might as well do.

  9. Doctor Syntax Silver badge

    How about increasing the throughput by using smaller drives and more of them, throwing in a few spares at the same time? We could call it Redundant Array of Inexpensive Drives.

  10. Stuart Halliday

    "How about a double pivot design? "The real question is will a 20GB drive with 2 positioners be enough cheaper, when cost of slot, expander port, etc., are factored in, than two 10TB drives to make it worth while.""

    Hehe... You have to laugh.

  11. DCFusor

    Race conditions galore

    Is what I forsee, especially when files are used for locking or are shared.

    One head set could be reading stale data while another set could be overwriting it with something else - which might not even be the same file in today's filesystem allocation schemes.

    Therefore there would have to be a layer keeping track of all this to ensure that a write queued up for one set of heads wasn't going to mess with a read in the que for the other set.

    On the head per track thing...been there, done that. Waaaay back in the day there were "drum memories", they were fast as crap - and hideously expensive compared to other means of storage even then. I fixed a few as a DEC field service guy back then, and have an old mil-spec one in my junkpile. The idea died out for good reasons.

  12. Jim O'Reilly
    Holmes

    SSD wins! Who cares if HDDs have 16 actuators?!

    I've heard all this before. IBM tried dual actuators, and so did Hitachi. There's a reason they are all history. Multi-actuator drives are expensive to make and not as reliable (vibration etc.). There have been attempts to make drives with even more actuators (drums) also ancient history.

    It's all academic anyway. By the time these drives hit Main Street, QLC SSDs, 3D NAND and a glut of dies will make HDDs obsolete altogether Sure, SSD drives will still cost more, but only maybe 2x, but 2.5" or M2 ruler form factors,coupled with 32-64 TB capacities will reduce appliance counts by as much as 4x for a given capacity...heck, Intel has announced 1 petabyte in 1U using ruler drives. Overall, SSD has to be cheaper!

    1. Anonymous Coward
      Anonymous Coward

      Re: SSD wins! Who cares if HDDs have 16 actuators?!

      Solid state has been on the verge of winning over spinning rust for the last 5+ years by my reckoning. I'm still waiting. Try building a 4x4TB NAS and tell me SSD is "Overall, SSD has to be cheaper!"

    2. katrinab Silver badge

      Re: SSD wins! Who cares if HDDs have 16 actuators?!

      I bought a 6TB WD Red for £180 earlier this week. Where can I get 6TB of SSD for less than that?

      1. Jim O'Reilly
        Holmes

        Re: SSD wins! Who cares if HDDs have 16 actuators?!

        If you were buying a car would you expect to pay KIA prices for a Ferrari?

        That's the comparison you just made.

        In the consumer market, you should be buying 256 GB to 1 TB SSDs as your primary desktop storage. The performance difference is really noticeable. Bulk storage should either be in the cloud or on the slowest drive around. Dual actuator drives don't make sense. Use compression on your slow drive...1TB raw capacity will hold around 6 TB of data.and performance will also be faster.

        1. Charles 9

          Re: SSD wins! Who cares if HDDs have 16 actuators?!

          Depends on what your slow drive is carrying. If it's stuff like media that's already compressed, best leave it alone.

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

Other stories you might like