Connect with us

Pan platter heads for hard disks

Discussion in 'Electronic Design' started by Siddhartha Jain, Nov 26, 2004.

Scroll to continue with content
  1. Hey,

    Me and my colleague were discussing the new PCI-Express architecture
    and sighed at how HDDs still lagged far behind other peripheral
    devices. But we both were curious about the same thing, that is, why
    can't HDDs be made with a head platter over a magnetic media platter?

    So instead of one head that moves over the HDD platter's surface, what
    if there was a single big circular head platter for each surface
    consisting of several heads? That would make mechanical movement in the
    HDD redundant, seek times will be zero and throughput would improve.
    Ofcourse, the costs would be much higher to integrate the extra
    components and the size could also grow but there are several
    applications that would greatly benefit from zero seek times and there
    are several IT deptts that will pay a hefty premium to get such a HDD.
    So is just tooooo expensive or impossible to engineer?

    - Siddhartha
     
  2. Whoo boy. You gonna run this thing in a hard vacuum?
    Been there, done that, since the dawn of computers, don't do it anymore.
    Search under 'drum memory', 'head per track', 'fixed head'.

    "Those that do not learn from history are doomed to repeat it."
    Somebody Famous - and he wasn't the
    first to say it

    The answer to the question 'has anybody ...' is 'yes'.
     
  3. keith

    keith Guest

    Have you researched current drives?
    How are you going to get the pitch of your "heads" to be the same as the
    pitch of the tracks? (Hint: HDDs use GMR heads, where 'G' = Giant) Even
    stepping around this little problem, how do you align your "heads" with
    the track. Current drives use the track itself as a servo feedback
    mechanism to center the head on the track. If your "heads" can't move,
    how do they center on the track?

    Ok, you've solved these trivial problems... Now consider that
    track-to-track head movement on current drives takes about the same
    time as a head switch (still have to re-acquire track center on head
    switch). What does it save you to have multiple heads on a surface?
    You want zero seek time? Buy solid-state drives.
    The evidence says you're wrong. There are solid-state drives, yet thay
    haven't taken over the IT world. Yes, they are expen$ive, but they
    *should* be everywhere if you were right.
    Soo many assumptions, soo little engineeeering. But, you did get one part
    right. Who wants to pay even a nanodollar per byte?
     
  4. Ian Stirling

    Ian Stirling Guest

    And on vibration isolating mounts.
    Let's not mention for the moment the horrendus problems you'd have
    with making bearings that are up to the task.
    Seek time would drop by about half, not be zero.

    It's important to note that track numbers have gotten insane.

    A 60Mb disk can be read out in a couple of minutes, and has maybe 5000 tracks,
    or a track spacing of maybe .00254mm
    A 60Gb disk packs the data in 30 times as much in each direction, so
    it takes around an hour to read, and has a track spacing of some .0001mm.
     
  5. http://www.dbazine.com/ault6.shtml

    Looks interesting. Seems like ten times more expensive but we all know
    what economies of scale can do.
     
  6. Ian Stirling

    Ian Stirling Guest

    Err, no.
    1Gb of flash is more like 30-40 times more expensive than 1Gb of hard
    disk.

    Solid state drives have been going to take over soon for a while now.

    I first read articles on this when a 20M PCMCIA drive was
    state-of-the-art.

    Now you can get drives a couple of thousand times bigger in the same
    format, but coincidentally, IIRC the price differential is
    pretty much the same.
     
  7. Depends on whether the units are RAID, and what type as well.

    --
    Dirk

    The Consensus:-
    The political party for the new millenium
    http://www.theconsensus.org
     
  8. Robert Baer

    Robert Baer Guest

    Sounds like the idea was to have multiple *fixed* heads (via a
    platter).
    Then, by definition, the heads create and read their own tracks; no
    servo needed or desired.

    Like someone else mentioned, drum memories had multiple fixed heads.
    And in the 80's, there was a company that made removeable hard drives
    that had multiple (moving) read arms for faster access. The drives were
    "bare" disk assemblies that went into the case that had the heads.
    Cleaniness problems made them die fast (3-6 month MTBF).
     
  9. It would still be needed to compensate for thermal expansion shifting
    the tracks, bearing slop, etc.
     
  10. nospam

    nospam Guest

    Trouble is heads are currently 3 or 4 thousand times wider than the tracks
    they read. So instead of one head reading 100,000 tracks you might have 100
    heads reading err, 100 tracks.

    Given the huge reduction in capacity semiconductor memory would be cheaper,
    faster, and more reliable.
     
  11. Rich Grise

    Rich Grise Guest

    Found an interesting article on the first drives with some stuff about
    fixed heads.
    http://www.mdhc.scu.edu/100th/Progress/Hoagland/hoagland1.pdf
    I had just looked up "Head-per-track" on google.

    I used to work at Control Data Magnetic Peripherals where they made
    drives. One of them had a "fixed head feature" that used a "shoe" with 32
    or 48 or so heads attached. At the time, they were all flying heads, but
    all that took was to mount them on springy arms. It was a pretty cool
    drive system - about the footprint of a desk, but almost chest-high, it
    had two removable 300 MB(!) packs of 10, 14" disks, with the shoe on the
    bottom, if they ordered it. I remember that that feature almost doubled
    the price of the pack. It was voice coil positioned - the voice coil was
    about 4" in diameter, and the magnet was about the size of a bowling ball.
    This was also where I took a class in the "Enhanced controller-tester,"
    which was a 74181-based, 16-bit (or maybe 24 - it was a long time ago)
    microprogrammable processor. Instruction cycle time 163 ns or something
    mind-boggling at the time.

    It was a pretty cool job. :)

    Cheers!
    Rich
     
  12. More like one big continuous head covering the entire surface area of
    platter's surface. Depending on the track/sector to be read, the head's
    corresponding track/sector could be *lit* up. So no moving parts at
    all.

    But then I zero knowledge of electronics. Am just a plain Unix admin
    who's tired of seeing capacities increase but HDD throughputs lag far
    behind.

    - Siddhartha
     
  13. More like one big continuous head covering the entire surface area of
    platter's surface. Depending on the track/sector to be read, the head's
    corresponding track/sector could be *lit* up. So no moving parts at
    all.

    But then I have zero knowledge of electronics. Am just a plain Unix
    admin
    who's tired of seeing capacities increase but HDD throughputs lag far
    behind.

    - Siddhartha
     
  14. In a sense, such magnetic memories exist (FRAM), price is the problem. The
    sheer number of 'cells' involved in a modern HD drive (1E12 is fairly
    'routine'), makes any solution that does not have the savings associated
    with being able to 're-use' the heads for a lot of cells un-economic for
    most applications. Even the idea of multiple heads (this was done years
    ago, on 'drum' drives), runs into the problems of the numbers involved (on
    a modern drive you can in some cases be talking in excess of 10000
    physical tracks...). In the past, 'bubble' memories looked for a while as
    if they would threaten the HD dominance, but the rate of capacity growth
    in the HD's, ousted this technology too.
    HD's, actually 'led the way' for many years, in the rate of improvements,
    relative to other parts of the computer. However as a relatively more
    mature technology, the problem is that just as the semiconductors will in
    a few years time, they are hitting the physical limits of the method now.
    If you want to change the state of a domain faster, it ends up having to
    be larger. So there is a balancing act between high rotational rates (and
    low latency), versus capacity. There are some technologies that were tried
    in the past, that may well make a return (multiple seperate R/W heads
    round the platter, to reduce the rotational latency), but since the
    biggest marketplace is largely 'cost driven', the pressure at the moment
    is just to pack more into a given surface area, rather than increase
    performance....

    Best Wishes
     
  15. keith

    keith Guest

    Nope, won't work. There is a reason the tracks themselves are used as the
    servo encoder. A few generations ago there was a special "servo track"
    and all the heads used that track for the servo invormation. However the
    track densities have increased to the point where even the dynamic errors
    *between* heads is greater than the track width. The answer is to have
    the tracks themselves act as the servo encoder. Thus a track acquisition
    is required even when switching heads on the same cylinder. One
    side-effect of this is that one can no longer predict the physical layout
    of a drive.
    Not at these densities!
    There have been "paging store" devices with four sets of heads at 90degree
    ofsets so that the rotational latency is dropped by 3/4. *Very*
    expensive, and no longer made. There are many better solutions. This
    one's a no-fly.
     
Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day

-