Nagered hard drive;'(..

Discussion in 'Computer Information' started by tony sayer, Jun 22, 2010.

  1. tony sayer

    D.M.Chapman Guest

    It's interesting, but I've dealt with all sorts of disk arrays over the
    last 15 years or so and it seems that while this becomes true of midrange
    kit as you get to the higher end, spare replacement becomes less urgent

    DotHill/Sun 3510 arrays - replacement disks sent to site within 4 hours.

    HDS AMS/USPs - disks sent next working day (or even slower).

    Admittedly, in our large enterprise arrays we do have many roaming spares
    (and often RAID 6) so I guess it's just deemed less important. Can't say
    it's ever been a problem though. No point sending a disk within hours
    when the machine spares out without problem and still has plenty of

    Controller failures however, they tend to respond to those pretty
    quickly ;-)

    D.M.Chapman, Jun 25, 2010
    1. Advertisements

  2. tony sayer

    Tim Streater Guest

    Yeah, OK, that's pretty common. I guess I never heard it called that
    before. But the I never used any ICL kit.
    Tim Streater, Jun 25, 2010
    1. Advertisements

  3. tony sayer

    Huge Guest

    Ooh, big iron! Mine was just an 11/23+
    Huge, Jun 25, 2010
  4. tony sayer

    Huge Guest

    Umm, the backup RAID needs to be in your DR site ...
    Huge, Jun 25, 2010
  5. tony sayer

    D.M.Chapman Guest

    Not always...

    Our offsite rep is much lower spec to save money. Decision might be that
    in a true DR situation with total loss of our primary DC that getting a
    few key services up quickly would be important - the rest could wait.

    A full same spec replica wasn't deemed cost effective. Not that uncommon.

    D.M.Chapman, Jun 25, 2010
  6. tony sayer

    Bob Eager Guest

    There's obvioously more, if it is to be useful, but that's a good start.

    We had a Honeywell DDP-516 that had a two-mode operation, but it was
    useless as (for example) there was no way of telling the previous state
    when an interrupt had occurred, so you couldn't restore state. Quite a
    lot of other holes too. I rewired the CPU to fix the variouls issues.
    Bob Eager, Jun 25, 2010
  7. tony sayer

    Tim Streater Guest

    Blimey wot was the point of that then (designing it that way, I mean).

    You prolly want to be able to designate memory as no-access, read-only,
    read/write, and execute-only, too (as well as mapping it).

    I remember going to a presentation on the then-new 68000 in 1979 where
    folks were asking about such features (and also what you mentioned), but
    the Motorola guy said that they figured it would take too much space on
    the chip. Turns out that later, when they looked into it, it didn't add
    much extra at all.
    Tim Streater, Jun 26, 2010
  8. tony sayer

    Bob Eager Guest

    It was only a little 16 bit machine, predating the PDP-11. Thousands of
    them were node processors on ARPANet.

    No idea why they did it like that...but history is littered with half
    baked solutions. I would guess cost. The 386 didn't correctly trap
    certain instructions, it just made them no-ops...which means hardware
    virtualisation wasn't possible. That fed through up to Pentiums quite
    recently. Now there's yet another operating mode to select to make it
    work as it should have done.

    (this rather ignores the fact that partial software virtualisation can
    actually be more efficient anyway!)
    Bob Eager, Jun 26, 2010
  9. tony sayer

    Phil W Lee Guest

    Total was about 30 employees doing fairly typical office work, so not
    that much data churn.
    It was OK once I managed to pry the boss's outlook data file out of
    the main data set - replicating 750Mb every time he fetched new mail
    was a bit of a bandwidth hog. :)
    So I put it on a different (non-mirrored) part of the disk and copied
    it to the DFS every night (before the backup run), and it was ok after

    But I have worked where the requirements were far higher than that -
    dedicated fibres linking the arrays were used there (but that was C&W,
    so they could afford it). That system wasn't my responsibility, so I
    don't know the full details, but the bunch of fibres going out through
    the basement wall from that array was about 3" thick, so I think it's
    safe to assume that was some major bandwidth. That RAID filled a 47u
    rack, so the principle is scaleable.
    Phil W Lee, Jun 26, 2010
  10. tony sayer

    Roland Perry Guest

    Inside there's a couple of large loops, technically it's a coil, but
    that's a confusing way to describe it.
    Roland Perry, Jun 26, 2010
  11. tony sayer

    Tim Ward Guest

    OTOH ... once upon a time a hardware designer designed a "useful" hardware
    feature which he thought would aid the programmers but actually made things
    vastly more difficult. I persuaded him to change to a much simpler hardware
    design that made the software easier, smaller and faster. So some
    communication between the two helps sometimes.
    Tim Ward, Jun 26, 2010
  12. Ah, but the CPU was the whole box. The OCP was the 'mill'.
    Rupert Moss-Eccardt, Jun 26, 2010
  13. tony sayer

    Bob Eager Guest

    Well, on our machine the three boxes were marked 'OCP'!
    Bob Eager, Jun 26, 2010
  14. tony sayer

    Steve Walker Guest

    Steve Walker, Jun 26, 2010
  15. tony sayer

    tony sayer Guest

    Did try that but most ever sector segment etc came up with unable to
    tony sayer, Jun 26, 2010
    1. Advertisements

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.