what types of long term data storage are used by modern mainframes?

Discussion in 'NZ Computing' started by Bling-Bling, Jun 11, 2005.

  1. Bling-Bling

    Bling-Bling Guest

    Just a question to help understand the differences between mainframes and
    other computers:

    What types of long term data storage are used by modern mainframes?

    Do they use the same HDDs as PCs? If yes, how are they connected?


    Bling Bling
     
    Bling-Bling, Jun 11, 2005
    #1
    1. Advertisements

  2. Bling-Bling

    Crash Guest

    There is no such thing as a 'modern mainframe' that can be described in
    generic terms. All (AFAIK) are proprietary and about all that could be
    accurately stated in that they use disk storage of some sort.

    The mainframes I am involved with have several entry-level models that
    are windows boxes - the mainframe environment runs as a windows app and
    the file system appears to windows as one or more big files. The
    character set used by the mainframe is EBCDIC. The high-end models use
    proprietary processors and EMC disk arrays - my employer no longer
    manufactures disk drives IIRC.

    Crash.
     
    Crash, Jun 12, 2005
    #2
    1. Advertisements

  3. Bling-Bling

    Bok Guest

    My guess is you work for UNISYS supporting MCP ClearPath machines.

    EMC provide storage solutions for IBM & UNISYS mainframes as well as
    SANs for high end 'open system' servers, oh and of course the UNISYS
    ES7000 server line.
     
    Bok, Jun 12, 2005
    #3
  4. It's the same basic disk drive hardware--there's no special "mainframe"
    type. The difference with the connection method is that it offloads most
    I/O processing from the CPU.

    Long ago, mainframes pioneered the concept of "channels", or what we now
    know as "DMA controllers" and are commonplace even on the cheapest PCs.
    Mainframes, in turn, have moved on to the next level, called a "Storage
    Area Network" (SAN), which is a proprietary network, using interfaces
    like Fiber Channel, which are far more expensive than Ethernet, for
    communicating with the disk drive controllers.

    But SAN is gradually being undercut by NAS (Network Attached Storage),
    which is the same concept, only using a cheaper Ethernet network. One
    protocol used with these networks is "iSCSI", which actually implements
    SCSI controller protocols on top of your common-or-garden TCP/IP
    protocol stack, so it can work over any hardware that already implements
    TCP/IP, i.e. a lot.

    As usual, the mass market comes in at the low end, nibbles around the
    edges, and eventually takes over the niche market.
     
    Lawrence D’Oliveiro, Jun 12, 2005
    #4
  5. Bling-Bling

    Don Hills Guest

    A little simplistic - mainframe "I/O channels" were complete processors in
    their own right. The main CPU would construct a "Control Block" in memory
    and transfer the address of the CB to the I/O Channel Processor. The IOCP
    would then do all of the initiation, data transfer and cleanup and post the
    ending status in the CB.
    In a PC, the CPU has to do the initial initialisation and command sequence
    to the I/O device and then it hands over to the DMA controller to do the
    data transfer, then the CPU does the cleanup at the end of the operation.

    Also, in a PC if the CPU wants to do another I/O operation while the first
    one is in progress it usually has to wait for the first one to finish.
    In mainframes, the CPU simply "daisy-chained" the new operation's CB to the
    current CB and the IOCP processed them in sequence. This removal of I/O
    overhead and waits from the main CPU is one of the reasons why mainframes,
    which traditionally had relatively low-powered CPUs, were able to handle
    large multi-tasking/multi-user loads, unlike most PCs which are still
    designed for relatively few simultaneous tasks.
     
    Don Hills, Jun 12, 2005
    #5
  6. Bling-Bling

    Jerry Guest

    Well into the 70's the mainframe channels were in separate boxes,
    measuring about 1 x 2 x 2 meters. Each held 1, 2 or 3 channels and had
    lots of flashing lights. There is a pic here
    http://www-03.ibm.com/ibm/history/exhibits/mainframe/mainframe_PP3165.html
    of a 370-168. The channels are the 3 boxes in the rear on the left,
    processor is the big cluster in the centre rear, the box on the right is
    the coolant (water) and power distribution unit. The operators console
    is just in front of the processor complex. Of course today the size of
    the channels has shrunk considerably, but the principle is the same.
    The channels work independent of the CPU after being given the initial
    start command for an IO operation. The Channels have direct access tom
    memory, and communicate with control units for the I/O devices (like
    disks). One computer can have many channels, and each channel can
    communicate with a number of control units, each of which can have a
    number of devices attached to it.
     
    Jerry, Jun 12, 2005
    #6
  7. Also the mainframe design is _not_ oriented towards notifying the CPU as
    soon as possible when an I/O operation completes--instead, it's
    optimized for maximum throughput, as opposed to minimum response time.
    Mainframes were, and as far as I know still are, designed primarily for
    batch, not interactive, processing.
     
    Lawrence D¹Oliveiro, Jun 12, 2005
    #7
  8. Bling-Bling

    Jerry Guest

    From my experience, which is primarily on IBM mainframes, the I/O
    device certainly notifies the channel, which notifies the processor as
    soon as an operation is completed. In most large environments there
    will be several systems sharing I/O, especially disks. The intelligence
    over the years has moved further from the system, resulting in more
    intelligent disk controllers and intermediate devices to handle what the
    system(s) want to do.

    Certainly batch work is still done on mainframes, but the move has been
    on interactive since the late (even mid)60's. Take a banking
    environment, payroll and scheduled transfers of funds are done in batch
    mode overnight, during the day the system must handle branch queries, as
    well as EFTPOS and ATN requests.
     
    Jerry, Jun 19, 2005
    #8
  9. Bling-Bling

    Bling-Bling Guest

    For decades New Zealand had only one realtime banking system - that
    operated by the Post Office Savings Bank.

    The system used by the then four trading banks - Databank, now EDS - was
    never and still is not a realtime banking system.

    For the trading banks, deposits and withdrawals are not actually
    processed as they occur. They are still processed overnight along with all
    other transactions. That banking system has only sufficient smarts to
    enable EFTPOS and AT machines to not overdraw the previous night's ledger
    balance and to take into account that day's held amounts. Even the tellers
    until very recently needed to take into account that day's held amounts
    when processing withdrawals across the counter.

    (Of course their newest terminal system now doesn't even require tellers
    to be able to add two numbers together!)

    On the other hand, the old system operated by the POSB was capable of
    having a deposit made in Oamaru, and the account holder withdrawing that
    money ten seconds later in Whangarei, provided both branches had working
    terminals. And the Account Holder's passbook would have been automatically
    updated with details of the deposit at the same time.

    I believe that the only other banking system that was similarly
    functioning in realtime was the old system used by the Auckland Savings
    Bank.


    Bling Bling
     
    Bling-Bling, Jun 19, 2005
    #9
  10. Then the Trustebanks one was in realtime
    When I make a transfer from one account to the other
    It shows the changed account balances immediately in the online
    banking
     
    FreedomChooser, Jun 21, 2005
    #10
  11. Bling-Bling

    Murray Symon Guest

    Out of curiosity, how many decades are we talking about here?
    ASB Bank and TSB Bank had real-time online banking systems in the 1980's.
    The original Databank/BNZ computer system began about 1967 - 1969.
    The Post Office system dated from about 1976.
     
    Murray Symon, Jun 22, 2005
    #11
  12. I don't remember the early databank being at all real time. Big batches of
    stuff come to mind.



    Bruce


    -------------------------------------
    The power of accurate observation is commonly called cynicism by those who have not got it.
    - George Bernard Shaw
    Cynic, n: a blackguard whose faulty vision sees things as they are, not as they ought to be.
    - Ambrose Bierce

    Caution ===== followups may have been changed to relevant groups
    (if there were any)
     
    Bruce Sinclair, Jun 22, 2005
    #12
  13. I was going to say that... I worked for Unisys until May/05... 16 years with
    them, and back in 2004 (when I started working with mainframes), the whole
    thing was proprietary - hardware, software, peripherals.

    I remember those disks the size of a washing machine - all 200MB of them...

    Now, look at this - my MS MVP award is for "mobile devices".
     
    Mauricio Freitas, Jun 22, 2005
    #13
  14. Bling-Bling

    Murray Symon Guest

    That's right, Bruce. That was just to establish a timeframe.
     
    Murray Symon, Jun 23, 2005
    #14
  15. Bling-Bling

    Bok Guest

    I had a couple of 3 year stretches with Burroughs corp before they
    merged with Sperry to become Unisys in 1986. First 3 years was spent
    supporting mainframe hardware, then I spent a year as a sysprog at
    Burrough's site then 3 years supporting System & environmental software.
    Stretching my memory a bit sounds like the capacity of a 207 Fixed disk
    drive.
     
    Bok, Jun 23, 2005
    #15
  16. Bling-Bling

    Bok Guest

    That's correct, they both had Burrough's mainframes back in the '80s and
    later Unisys ClearPath boxes. Their main online banking system were
    developed in LINC - recently TSB converted their LINC system to JADE
    deployed on IBM server hardware.
    Yes, that sounds about right.
     
    Bok, Jun 23, 2005
    #16
  17. Bling-Bling

    Chris Hope Guest

    They aren't processed in realtime? So how do you explain one of my
    customers transferring money into my account from their account at
    around 11am this morning, and it then appearing in my account
    immediately? Or me transferring money from one account onto my visa and
    the balances of both being updated immediately?

    [snip]
     
    Chris Hope, Jun 23, 2005
    #17
  18. Bling-Bling

    Bok Guest

    That was also my experience with non IBM mainframes. Some architectures
    had an intermediate I/O processor between the CPU and channel (or
    equivalent). There was considarable varation between hardware families
    (machine architectures) in the detail of how this worked but generally a
    significant event such as seek complete (disk heads moved to the correct
    cylinder) was notified via a disk controller to the "channel" to
    complete the data transfer part of the operation. Once the data was
    transferred to/from a memory buffer, then the CPU was notified via a
    (maskable) intterupt.
     
    Bok, Jun 23, 2005
    #18
  19. Yep, talking about the 207 disks...
     
    Mauricio Freitas, Jun 24, 2005
    #19
  20. Bling-Bling

    Murray Symon Guest

    Yes, that's true - the changeover took place on 09/05/2004, to be precise.
     
    Murray Symon, Jun 24, 2005
    #20
    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.