Notes 8 --- Other Secondary Memory Devices
(Claybrook, Chapter 2; Miller, Chapter 4 on PC disks; plus supplements --- see also BYTE Magazine, March
1994, Special Section on Memory Hierarchy, 78--116)
PC Disks
These include
- single-surface 3.5 and 5.25 inch disks
- semi-permanently mounted hard disks and hard cards
- high-storage disk packs like Bernouilli boxes
- removable hard disk packs used to support laptop-to-desktop data movement.
For the most part, they can be considered as equivalent to the disks discussed in Notes 7. Some other remarks
follow.
- We're all familiar with 3.5 and 5.25 "floppy" disks. These disks invariably have exactly one platter, and
usually two surfaces.
- The advantages of these truly portable disks are well-known to us all, but the fact that the disk surfaces are
not truly fixed in position relative to the read-write heads adds an extra factor to the access time; namely, the time
it takes the heads to move into proximity to the disk surface (head settle time).
- If the head were kept in this position permanently, there would be a serious risk of its being damaged as disks
were inserted and removed from their drives.
- PC hard disks, such as IDE and SCSI hard disks, are like the disks we have discussed in class, but come
packaged with their own controllers, and often extra features such as an external cache.
- Bernouilli boxes and transportable disks act somewhat like the Winchester boxes described in Miller.
- Hard cards come in two flavors:
- In one approach, the card is only a support for a hard disk, and allows easier installation. I believe that most
hard cards are of this form.
- A true hard card would be instead a two-dimensional rectangular storage device.
- Such devices could not, of course, rotate, but will have to use one of the following:
- a read-write arm with two degrees of freedom.
- a line of read-write heads spanning one card dimension, with movement in the perpendicular dimension.
- electronic access, like large core stores (see below).
- Their main advantage is in space but there are compensating problems:
- in the first strategy, each access to a new block will require movement of the head
- the second has most of the disadvantages of a fixed disk with none of the advantages.
- the third will be more expensive than a read-write head technology.
- It looked for a while as if these would replace or compete with hard disks, but one doesn't hear much about
them anymore, possibly because of CDs.
Large Core Stores
Definition:
- Large Core Store (LCS) is basically an electronic extension of main memory.
- It is not, however, based on semiconductor technology, but on the older magnetized core storage (responsible
for the dreaded phrase "Core dumped").
- It has properties intermediate between primary and secondary memory.
Properties:
- Unlike all other secondary memory devices, it has no moving parts, and thus no latency or seek time.
- It is typically addressable at byte or word level.
- It may be accessible from the CPU, but is more typically managed either by a controller, or by an indexed
access from the CPU, or as virtual memory (viewed as part of main memory via a page table).
- It is significantly slower than main memory, but faster than other secondary memory.
- It is persistent.
- It is significantly more expensive than most other secondary devices.
In systems in which it is used, it is typically used for storing system information which
- is needed quickly.
- is either going to be
- used on a regular basis (and needs persistence), such as compilers, or
- accessed from or swapped into main memory frequently, such as systems tables, file indices, space for
swapped-out jobs, or
- used for system update and interrupt handling.
- Claybrook suggests that its most important role may be as an external cache.
The mechanics of LCS will be discussed in class.
Charge-Coupled Devices and Magnetic Bubble Memories:
Charge-coupled devices (CCDs) and magnetic bubble memories (MBMs) can also be used to fill the gap between
semiconductor memory and disks. Each, in some sense, can be thought of as the secondary memory equivalent of a
flip-flop or shift register.
- A CCD is a buffer consisting of a linear/circular array (typically of size 16--128K) of electronic elements,
each capable of propagating one bit of information, actually laid out in two dimensions.
- The various types of CCD differ primarily, other than in size, in the 2D layout. [Figure to be added.]
- There is a refresh, and a single input and output position. You can think of this as providing the equivalent of
a single track of a disk.
- Input has to be synchronized with available space; output has to be read when the appropriate bits are passing
the output gate.
- Clearly, input and output can be managed at word or byte level, not just in terms of blocks.
- As in LCS, there are no physical "moving parts", but unlike LCS, the data moves, and is not in the same
physical location at all times.
- The technology is stable, but relatively expensive. Access is fast but scales at best linearly (doubling the
storage will double the access time). It is being used, but primarily in specialized applications.
- MBMs provide a hierarchical alternative to CCDs.
- MBMs are based on the N/S magnetization of domains in a magnetic field. With the right kind of field, we
get true binary magnetization, and the magnetic domains "drift" --- although the underlying physical substrate does
not --- providing the same sort of rotation as in CCDs or disks. One can set up a number of such "rings" of
domains in the same substrate.
- Hierarchical organization is achieved by making transfer of information between rings possible; however,
each such transfer point requires another controller element.
- Unlike CCDs, I haven't heard anything of MBMs recently. I suspect the technology is at best still
experimental. However, work in high-temperature superconductors may make these a more reasonable choice in
the future.
Optical devices:
Perhaps the most truly astonishing development in secondary storage has been optical storage technology.
Claybrook, published in 1983, doesn't even mention them, and for that matter, neither does Miller (although she is
not aiming for thorough coverage).
The current state of the art can be summarized as:
- Read-only (ROM) or write-once (WORM) disk technology is easy, and CD-ROM disks are widely used. The
storage capacities are truly immense, and access time is still comparable to disk access time. Supporting hardware
is more expensive than for disks, but not overwhelming.
- CD-ROM technology typically uses only a single platter and surface.
- Read-write disk technology is not as well established, a combination of a somewhat-still-experimental nature
and a high price for both hardware (the controllers, etc.) and media (the disks themselves).
- Current rewritable audio-disks use a different laser technology. While the storage medium is denser than that
used for read-only disks, there are some problems with fidelity, which would make this a poor choice for secondary
storage.
- More experimental technology (3D disks, holographic disks, etc.) are in development, or being used as
prototypes.
- The technology for CD-ROM involves pitting the surface of the disk --- a given location is either pitted or not.
It's easy to see why this makes reading a good deal easier and less hardware intensive than writing. It's also not
hard to see why extension to "write-once" (WORM) is not that difficult --- after all, it's filling existing pits that
really causes the difficulty.
- Reading a CD-ROM involves a laser which shines on a cell; if the cell is unpitted, we'll get a specular
reflection, while if it is pitted, the reflection will be diffuse.
- Read-write optical disks have to use a completely different (although still laser-based) technology, relying on
media which change their chemical or physical properties when excited by laser light (and stay that way! ---
without this requirement, things are much easier).
- One conceivable technology might be to shine a beam onto a cell which has been rendered opaque by one kind
of beam, or transparent by another, and have a detector below determine whether the beam passes through the cell.
- A more likely technology (and I think the one that is being used) is to use two beams: the write beam changes
the polarization of the cell; the read beam passes through a cell polarized as 1, but not through a cell polarized as
0; there is still a detector below the read beam.
- 3-D technology uses two beams for read and for write, one for the level (think of surfaces) and one for the
block. The block beam will always pass through any cell through which the level beam is not shining, so the
detector will read the state of the cell at the intersection of the beams.
- The difference between this and multi-platter disk technology is that this is really a 3-D medium, divided into
volume elements, rather than a set of plates, divided into surface elements.
Mass stores:
Essentially, a mass store is like a juke box. There is a set of tapes (or CDs, or, less frequently, single-platter disks
or disk boxes), plus a mechanism for selecting and mounting one of these on demand to a tape/CD/disk controller.
When a file is needed, the process is: if the correct tape/etc. is mounted, do nothing. Otherwise, unmount the old
tape and return it to its proper place; fetch the right tape and mount it. Then the tape controller takes over.
- These are now being used a lot for CDs, less for tapes, and far less for other media.
CD packs:
These closely resemble multi-surface hard disks, but each platter is in fact a removable CD. While in principle
both surfaces of the CD could easily be used in such an arrangement, in practice only one surface is used, since
that’s the structure of most CDs.
END