Text preview for : 19780707_A_New_Class_Of_IO_Device_The_OIS_Archive_Device.pdf part of xerox 19780707 A New Class Of IO Device The OIS Archive Device xerox sdd memos_1978 19780707_A_New_Class_Of_IO_Device_The_OIS_Archive_Device.pdf



Back to : 19780707_A_New_Class_Of_I | Home

XEtOX eDD ARCHIVES
1 haVi read and underBtoo4
XEROX
BUSINESS SYSTEMS
Pages _ _ _ _- To
Reviewer_ _--Da.t\jj-~-
---===-
Systems Development Department 'f: of pages Ref . .?a=,s lJjj -1.$1
. July 7, 1978


To: Dave Liddle

From: Peter Bishop / SD at Palo Alto

Subject: A new class of I/O device: the OIS archive device.

Copies: W. Lynch, J. Weaver, J. Wick, S. Wallace, R. Sonderegger, R. Metcalfe, W. Kennedy, J.
White, V. Schwartz, R. Belleville, T. Townsend, J. Mendelson, R. Wickham, C. Irby, L. Bergsteinsson, W.
Bewley, J. Szelong, P. Heinrich, E. Harslem, L. Clark, D. DeSantis, J. Reiley, G. LeCesne, D. Reilly,
Dave TIlOrnburg, Brian Rosen, Bill Gunning


Introduction

Developments in LSI technology will have important effects on the cost of memory, and therefore upon
the architectures of computers in the 1980s. Many people who try to predict what computer architectures
will exist in the 1980s ignore the huge cost of developing software for drastically new architectures. This
note assumes that basic computer architectures will not change, i.e. that the DO will be a state-of-the-art
processor (in the low-moderate price range). It is relalively clear that in the 1980s there will be two major
factors in the cost of a computer system: peripherals and memory. Therefore the cost of memory can be
expected to affect the system architecture. A recent report by Mackintosh Consultants Co. Ltd. [1] has
taken a close look at the costs of memory in many different technologies. Using this report, we can
estimate what technologies we will be using for memory in the 1980s.

An important factor that cannot be ignored is the effect of price competition from new memory
technologies (such as bubble memory and CCDs) on the structure and price of old, established memory
technologies (such as disk). One conclusion of the Mackintosh report is that there is no future for fixed
head disk technology because by 1985 bubble memories and CCDs will be cheaper while offering a
faster access time than fixed head disks. Mackintosh further predicts that beginning in 1980 bubble and
CCD memories wUl be marginally cheaper than RAM, but by 1985 bubble memories will be one order
of magnitude cheaper than RAM. The 1985 cost of bubble memories, however, will still be twice as
expensive as current costs for non-removable moving head disks. Mackintosh expects the cost of non-
removable moving head disks to drop to one-third of their present cost by 1980.

An interesting point made by the Mackintosh report is that it is questionable whether there will be a
place for removable disks in the 1980s. The main reason for this is tiwt a removable disk cannot be as
tightly sealed as a non-removable disk. A tightly sealed disk can operate with closer tolerances that
reduce noise and allow higher densilies to be achieved on tile disk. 'TIms alUlOugh the cost per bit of
non-removable disks will fall, it is not clear that the cost per bit of removable disks will also fall. A major
advantage of disk memory is that it can be used as a pseudo random access device for the on-line storage
2

of infomJation. Cartlidge disks anow the on-line information to be removed, but also to be quickly
placed on-line on another system. As the relative cost difference between removable disk and non-
removable disk increases, other methods of moving information from system to system and other
methods of achieving off-line storage may be found.

We at PARC have already found that high-speed communications lines may be the best way to move
information from one computer to another. When wc begin to look at technologies that can achieve
efficient ofT-line storage, we notice that magnetic tape achieves an extremely low cost-per-bit for off-line
storage. Many people have noticed that magnetic tape is hard to handle, but the tape industry has
answered this objection by developing tape cassettes. Tape cassettes are just as easy to handle as cartlidge
disks, but achieve a much higher surface area. An exciting prospect in the early days of integrated circuits
was the possibility of producing three-dimensional circuits. Although this has not been realized in
integrated circuit technology, magnetic tape realizes this ideal to a surprising degree.

Use of an Archive Device

There are basically two uses for an archive device: archive and backup. When used for backup, there
would be one archive-device volume that contains the last complete dump of the system and all of the
changes that have been made to the system since then. If the capacity of the archive volumes is not
enough for this, then dumps may be placed on several volumes. When the amount of storage used for
incremental backups becomes the same order of magnitude as the storage used for a complete dump,
then another complete dump is taken. To restore the system, it is necessary to re10ad the complete dump
and then replay the incremental dumps. The larger the capacity of an archive volume, the less frequently
the archive volume will have to be replaced by an operator. If the capacity of an archive volume is 3-10
times larger than the on-line storage of the system, it may not be necessary to ever replace the archive
volume, since an old dump can be overwlitten with a new dump. In addition, an archive device can be
used for archival storage. Reading or writing an off-line volume requires operator intervention to mount
the off-line volume. It is acceptable to use a device intended for backup for archival as well since reading
or wliting archive volumes will only use the device for a short time. The backup volume would be
dismounted, the archive volume mounted, the read or write performed, and then the backup volume
remounted.

High Performance Requirements

This suggests that there is a place for a special archive device that has an interestingly different behavior
from all of the classical digital storage devices. The basic requirements for this device are:

1) low cost for the device itself, since it is a special-purpose device.

2) extremely low cost per bit, especiaUy for off-line storage.

3) high transfer rate - it would be nice to be able to dump a disk onto the archive device and be
limited by the disk rather than the archive device. [It would also be interesting to be able to keep
a log of X-wire traffic (or an interesting subset of X-wire traffic).] This requires a transfer rate of
at least 106 ,bits per second, but transfer rates up to 107 bits per second would be nice.
" " ,

4) very large capacity