BRL 1961, ANALYSIS AND TRENDS, start page 1025
|
CHAPTER III
ANALYSIS AND TRENDS
BRL 1961, ANALYSIS AND TRENDS, start page 1026
|
ANALYSIS AND TRENDS
INTRODUCTION
The information for each of the 222 systems described in Chapter II has been subdivided into eighteen topics, permitting the data
to be presented in
an organized manner and simplifying the comparison of features of the different systems. The following paragraphs, paralleling the
subdivisions of the
systems descriptions of Chapter II, are an attempt to quantitatively analyze the data and show recent trends in the field of
computing machinery. It is
emphasized again that the information given in Tables II through XV in this Chapter is to be used with caution. The tables have
been constructed only to
show trends, permit limited comparison of systems and show the present state of the art. Information pertaining to a specific
system should be obtained from
the system description in Chapter II or directly from manufacturers and users.
DESIGNATION OF COMPUTING SYSTEMS
The names of various types of computing systems existing in the United States stem from different sources. It would have been
convenient if some
system of classification and standard nomenclature had been established many years ago. The nomenclature could have incorporated
the name of the
manufacturer and model number, the nature or application of the system, or the name or location of the operating agency. However, a
system of nomenclature
was not established, resulting in an odd mixture of names for computing systems. Many computing systems bear the name of the
manufacturing organization,
for example IBM 704, HONEYWELL 800, NATIONAL 304, ILLIAC, and RCA 501. The names of some machines indicate the nature or purpose of
the
system, for example WESTINGHOUSE AIRBORNE, VOTE TALLY SYSTEM, CUBIC AIR TRAFFIC, WHIRLWIND and EDVAC. Other machine titles
indicate the name of the operating agency, such as DYSEAC, SEAC, NORC, OARAC, ORACLE and ORDVAC. Some titles are indicative of the
location of
the system, such as LARC. The names of some machines are trade names like UNIVAC II and ET OM 125. There are some machines named
after specific persons,
as are ALWAC III E and JOHNNIAC. Arbitrary names, like GEORGE, also exist. Another trend in computing machine nomenclature has been
to develop
names which were contractions or pronouncable abbreviations of significant titles. Examples of this are EDVAC, for Electronic
Discrete Variable Automatic
Computer; MANIAC, for Mathematical Analyzer and Numerical Integrator And Computer; and ORDVAC, for ORDnance Variable Automatic
Computer.
MANUFACTURERS OF COMPUTING SYSTEMS
In the interest of national defense, the development of electronic computing systems could not wait until normal economic laws
brought about the
supply of systems through commercial demand. The Department of Defense supported research and development in the field of
electronic digital computers to
be utilized for rapid scientific computation on defense projects.
The world's first electronic digital computer, the ENIAC designed and developed by the Moore School of Electrical Engineering of
the University of
Pennsylvania, for the Ballistic Research Laboratories was placed in operation at the Aberdeen Proving Ground in January 1947. Many
early electronic
machines were manufactured at educational institutions such as the Institute for Advanced Study, MIT, Harvard and the Universities
of Pennsylvania and
California. Parallel research was performed by industry, and by 1950, large scale digital electronic computers were being delivered
commercially. At the
present time mass production of large scale systems is well underway. Several thousand large scale systems of various types have
been mass produced, and
thousands axe on order. Table I shows the manufacturers of all the machines described in Chapter II and Table II shows the
approximate quantities of these
systems which have been produced.
BRL 1961, ANALYSIS AND TRENDS, start page 1027
|
APPLICATIONS OF COMPUTING
SYSTEMS
The installation of the ENIAC, at the Ballistic Research Laboratories of the U. S. Army Ordnance Corps marked the
beginning of the widespread use of electronic computing machines. Since the advent of the ENIAC, a large expansion has taken
place in the computer field. Investment rates in computing equipment in the United States have risen from ten million dollars
per year in 1953 to one hundred million dollars per year in 1956. Present expenditures for computing equipment has passed
the billion dollars per year mark.
Almost every commodity industry such as oil, steel and rubber is utilizing computing equipment for both scientific and
commercial applications. Service industries, such as banking, transportation, and insurance have applied large scale systems
toward the solution of problems in the fields of accounting, reservations control, and bookkeeping. Manufacturers have used
computing systems for design engineering and scientific research. Many systems are being utilized for inventory and stock
control. The determination of manufacturing plant location and stock parts storage are being made by linear programming
methods. Electronic computers are being utilized by the construction industry for design and location of structures and road
nets. Many digital computers form a part of closed loop industrial process control systems.
Many problems require the processing of large quantities of data, such as is obtained from missile tracking,
telemetering, mineral deposit prospecting and record keeping. The use of electronic computing equipment permits the
processing of large quantities of such data over relatively short periods of time.
Many "on-line" applications of both general and special purpose computers are being made. These control
applications include such examples as control of wind tunnel testing and continuous-flow manufacturing. Computers are
being used for aircraft and missile fire and flight control, both as ground based and missile borne systems.
A discussion of applications of specific systems will be found under the sub-heading "APPLICATIONS" in the various
computing systems descriptions given in Chapter II.
PROGRAMMING AND NUMERICAL
SYSTEM
Internal Number System
Many types of number systems have been utilized for the development of logical designs of computing systems. Among
these number systems are the straight binary, octal, binary coded decimal, straight decimal, sexadecimal, biquinary, binary coded
alphanumeric, and binary coded decimal (excess three). Of 187 different relevant systems, 131 utilize a straight binary system
internally, whereas 53 utilize the decimal system (primarily binary coded decimal) and 3 systems utilize a binary coded
alphanumeric system of notation. Of course, in nearly every computing system, information is ultimately handled in binary form,
particularly in storage and in arithmetic units. The primary method of storage exploits the inherent properties of material media,
such as semiconductors, and ferroelectric and ferromagnetic materials. The state of conduction or the polarization of
ferroelectric and ferromagnetic materials determine the nature of the information which is stored or being processed. Decimal
digits are handled as groups of four bits, or tetrads. Alphanumeric data usually requires the use of six bits, permitting 64
different
symbols. Some systems utilize seven bits for expressing a single character, permitting 128 different characters, or may utilize a
single bit as an "odd-even" check bit. Programmers and coders preparing problems for solution on these systems may work
with decimal or alphanumeric notation and need not be concerned with the binary coding performed automatically by the
machine.
Word Length
The selection of word length for computing systems is based upon many considerations. For information words, the
precision required for the solution of problems may be the major consideration. For instruction words, word space must be
allocated to the address of the operand (or operands for multi
BRL 1961, ANALYSIS AND TRENDS, start page 1028
|
address codes), the command, and perhaps spares, tags, or check digits. For example, the ORDVAC utilizes 39 bits plus sign for an
information word. One-half
of a word, or 20 bits, is subdivided into a 12-bit address portion (for 4,096 high speed storage locations), a 6 bit command
portion for 64 commands, and a 2-
bit spare digit portion for special applications and versatility. The variation of word length among existing systems is rather
wide. Table III shows the word
lengths of the 222 systems described in Chapter II, in ascending order of magnitude. The average or nominal word length for fixed
word length machines is
approximately 40 binary or 12 decimal digits.
Number of Instructions Per Word
In many systems the machine word structure permits several instructions to be expressed by a single word. Of 171 systems,107 were
reported as
operating on a one instruction per word basis and 28 were reported as operating on a two instructions per word basis. Several
systems required two words
to express a complete instruction and, in some systems, several instructions could be expressed by a single word, at the option of
the programmer.
Arithmetic System
Most of the earlier machines operated on a fixed point arithmetic system. The binary or decimal point was arbitrarily fixed at
either the right or left
end of the number. For some systems a centered decimal point permitted the direct expression of whole and fractional parts of
numbers. Scaling is required,
for example, when a decimal or binary point is located at the left end of a number, in which case all quantities must be scaled
between the values of minus
one and plus one.
Many of the later machines were manufactured with built-in automatic floating point equipment, permitting numbers to be expressed
as fractional
parts and exponent parts. The exponent usually is a power of two or ten. Floating point circuitry was added to many of the older
systems. A review of this
sub-heading in the systems descriptions found in Chapter II and an examination of Table III will show the distribution of fixed and
floating point equipment.
Instruction Type
Internally programmed automatic computers require that part of the instruction word be devoted to the address (or addresses) of the
operand (or
operands). The question of how many addresses are to be incorporated into a single word has been answered in many ways. In single
address machines, the
address of one operand is given in the address portion of the instruction word. In two address machines, the address of two
operands are given, for instance
the addresses of the minuend and subtrahend are given for a subtract instruction. For three address machines, the address for
storing the result, e.g., the sum,
difference, product, quotient or square-root, is given. The three address machines usually refer automatically to the next storage
location, in sequence, for the
next three-address instruction word. Machines using the four-address instruction will express the location of two operands, the
location for storing the
results of the operation, and the location of the next instruction, all in one four-address word. In a 1 + 1 system of notation the
address of an operand for the
current instruction is given, along with the address of the next instruction to be performed. Coding for four-address machines is
somewhat simplified,
however, a more complex machine structure is necessary. The following table shows the distribution of different addressing systems
among the types of
computers described in the handbook.
Instruction Type Different Systems Using Given Type Instruction
One-address 116
Two-address 23
One or two-address (optional) 13
Three-address 20
Four-address 7
One-plus-one and one-over-one address 8
One and one-half address 3
One or three-address (optional) 2
BRL 1961, ANALYSIS AND TRENDS, start page 1029
|
One or one-plus-one address (optional) 2
One, two, or three-address (optional) 2
One, two, or four-address (optional) 1
Modified three-address 1
Three or four-address (optional) 1
Variable up to five-address (optional) 1
Total 200
Instruction Word Format
Most systems require adherence to a specific format or sets of formats for preparing coded instructions, in the machine language.
The instruction word
format thus outlines the form in which the instruction is prepared. An accounting must be made of each digit or character of the
instruction word.
ARITHMETIC UNITS
Operation Time
Since the primary function of an arithmetic unit in any computer is to perform repetitive arithmetic operations rapidly, the time
required to execute an
add instruction or a given sequence of arithmetic or logical instructions, is extremely important when selecting a computing system
for a specific application.
Tables IV and V were prepared to show at a glance the general state of the art with respect to arithmetic speeds. It must be
emphasized that the values stated in
the table are on an "as reported basis". The reader is reminded that the tables must be used with caution, since many clarifying or
related remarks have been
omitted for the sake of simplicity. Refer to the system descriptions of Chapter II for further detail.
Table IV shows the approximate relative order of add time when including the storage access time. In many systems, it is not
possible to determine the
time required for one addition without considering storage access. This may be due to the fact that in many types of operation,
sums may form in an
accumulator as the addend is brought from storage, hence access time may be inseparable from add time.
Construction of Arithmetic Units
Most of the computing systems described in this report utilize tubes or transistors as the basic driving element in the arithmetic
unit. Several
systems utilize magnetic cores in the arithmetic unit. Gating for arithmetic and logical units is most usually performed by diodes,
transistors, or vacuum
tubes. A review of the construction methods used in arithmetic units is discussed under this topic in the systems descriptions.
STORAGE
An extremely diverse and dynamic field of interest in the study of computing systems is the subject of storage devices. Many
ingenious devices,
utilizing the ability of various material media to store or record energy transformations, have been devised. Early forms of
storage involved mechanical
deformation of material media. These are exemplified by cams, springs, gears, music box cylinders, perforated player piano rolls,
code wheels and perforated
paper tape. All these storage devices required the movement of large masses of material and consequently long access time was
inherent. The capacity, in terms
of stored information per unit volume of material, was very low.
During World Wax II, the search for more rapid access storage devices led to the use of the vacuum tube. The two states, that of
conduction and that
of cut-off, permit information storage on a binary basis. This system, as was used on the ENIAC, proved effective from an access
time consideration,
however, the system was extremely bulky and required thousands of electronic vacuum tubes for a storage unit consisting of only 20
words of 10 decimal
digits each.
Chronologically, the next development was the use of acoustic delay lines of mercury and quartz. A transducer at each end of a
length of these materials
permits energy conversions and allows the storage of information in the form of high frequency (e.g. 8 megacycles/sec) pulse
packets. The information is
BRL 1961, ANALYSIS AND TRENDS, start page 1030
|
continuously recirculated. Information is inserted or read out through the use of standard gating techniques. Among the computers
utilizing acoustic
mercury delay lines are the DYSEAC, EDVAC, ELECOM 725, SEAC and UNIVAC I. Quartz acoustic delay lines were also used. Other types
of delay lines
used for storage of information are the magnetostrictive and the electromagnetic or distributed L-C network. See Tables VI, VII and
VIII, which list the
computing systems utilizing delay line storage units. Although in operating principle there is no difference, it is necessary to
make a distinction between a
delay line used in a storage loop in which information is continuously circulated, and a delay line used only for purposes of
timing the arrival of information
at selected points for performing various logical operations. In the latter, the function is delay, or temporary storage, rather
than permanent storage. Since
delay lines store information serially as a train of electrical or sonic pulses, average random access time is limited to half of
the time length of the delay line
plus the time equivalent to one word length. Because of the serial nature of the system, delay line storage units are limited in
speed. Notice how the delay
line types of systems lie near the bottom of the Access Time of High Speed Storage, Table VI.
The search for shorter access time brought about the development of the electrostatic storage unit, also called the cathode ray
tube storage device.
The material medium in motion was now limited to electrons, i.e., in beams and on charged areas on the screen of a cathode ray
tube. These charged areas
behaved somewhat like an array of charged capacitors. Selection of storage locations and the transfer of information was
efficiently performed by an easily
deflected pencil or beam of electrons which was used for both writing and interrogation. Parallel transfer, in which all digits of
a given word are transferred
simultaneously, became possible with this type of storage system.
The electrostatic storage system, with the inherent problems associated with high accelerating voltages, screen imperfections and
other tube failures,
has all but yielded to the utilization of magnetic cores for the storage of information. A 32 x 32 array of ferrite cores, which
might constitute a typical storage
plane, may measure only a few inches on each side. The cores are placed at the intersection of the wires of a mesh, and a third
winding may be threaded
through all the cores for sensing stored data. The storage takes place in the form of magnetically oriented molecular or atomic
dipoles which retain their
orientation upon removal of the magnetizing force. Many manufacturers intend to
provide computing systems with large capacity core storage units. Advances have been made in the use of perforated ferrite plates
and magnetic films
deposited on glass as a magnetic storage unit. Two such systems, the LINCOLN TX 2 and the UNIVAC 1107 utilize thin films. The
storage principle is the
same as for magnetic cores. Table VI shows the access time of high speed storage units in their approximate relative order of
magnitude for the storage units
used in various computing systems. It must be emphasized that the question of precisely what constitutes access time cannot easily
be resolved unless a
common understanding as to the definition is reached. In the usual sense, one may consider access time as the elapsed time between
the initiation of a
command to transfer an item of information, usually one word, from one address in the storage to another designated register, and
the complete arrival of the
item at the designated location. In many systems, particularly serial storage units, access time depends upon the time location of
the word in the serially
circulating group of words at the instant the transfer command is initiated. For this and other reasons, much misunderstanding can
arise in the consideration of
access time. the data presented in Table VI should therefore be considered to be approximate and should be used with caution.
The capacity of high speed storage units has risen during the past few years as rapidly as access time has diminished. Table VII
shows the capacity
of high speed storage units in terms of numbers of words and word lengths, arranged in relative order of magnitude of equivalent
binary capacity.
Rapid access storage of limited capacity is usually supported by a larger capacity storage unit for a well balanced storage system.
This permits the
transfer of large blocks of information from the rapid access storage unit to the large capacity storage unit for use at another
location or time in the
BRL 1961, ANALYSIS AND TRENDS, start page 1031
|
computation process. The most prevalent devices for auxiliary storage of this type are the magnetic drum or the magnetic disc. The
access time for large blocks
of information is of the order of tens of milliseconds for most magnetic drum or disc units. Many computing systems utilize
magnetic drums or discs as the
primary storage unit. Several systems utilize large capacity drum or disc units particularly for commercial type applications, such
as payroll, stock inventory,
and personnel records where access times of the order of microseconds are not required. Table IX shows the capacity of various drum
or disc storage systems
currently in use. It should be remembered, when glancing at Table IX, that although an attempt was made to show maximum capability,
additional drum or disc
units can be attached to some systems. Many systems employ magnetic tape as a medium of storage. Although access time is relatively
long because of its
inherently serial nature, a large vole of data can be stored on tape with a high packing density in terms of data units per unit
volume.
The characteristics of a storage device, namely, capacity and access time are two aspects of a storage system which come under
consideration when
designing or using a machine. The user or manufacturer of a system, at times, can trade capacity for access in the sense that under
certain conditions he can
accomplish an equivalent amount of computation with a large capacity, somewhat longer access time system as with a small capacity,
short access time system.
This is the old problem of trading time for space or vice versa. There are limits to this however, for example, when access time
approaches the order of
milliseconds, computation is seriously slowed down. Since large capacity and short access time are features to be desired, let us
examine a quantity determined
by the expression:
1og10 (Capacity in Equivalent Binary Digits/Access Time in Seconds)
In early storage devices, such as music boxes and signal coding equipment, this number is of the order of two to three. Relay
storage units have a number
of the order of four or five. Tube registers of the ENIAC vacuum tube accumulator storage type, enabled this figure to be as high
as 6.3. Magnetic drum
storage units are in the region of 6 to 7. Acoustic delay line storage systems show that this figure is in the range 8.6 to 9.6.
The cathode ray tube storage
(electrostatic) raised the figure as high as 10.79. The magnetic core storage unit permitted an increase of this figure to over 12.
Thin films have now arrived
d on the scene as a practical storage medium. The following table shows the growth, or increase of this number, as development of
computing system
components progressed:
Approx. Median Approximate Year
Storage Device Log10 Capacity/Access of Development
Early Mechanical 2 - 3 Prior to 1930
Electromechanical 4 - 5 1935
Vacuum Tube 5 - 6 1940
Magnetic Drum 6 - 7 1945
Electrostatic (CRT) 9 - 10 1950
Static Magnetic (Mag. Core) 9 - 12 1955
Thin Film 10 - ? 1960
Table VIII is a tabulation of the Log10 Capacity/Access figures for the high speed storage units of
various computing systems in approximate relative order of magnitude.
INPUT-OUTPUT
The above discussion on arithmetic units and storage devices have shown the great strides that have been made in these fields
during the past
several years. Arithmetic operation and storage access times have decreased and storage capacity increased. Yet, the communication
link between the
person and the machine still presents a major problem. Paper tape and cards, inherently bulky, are prevalent and relatively slow,
particularly for scientific
applications. The main convenience afforded by cards, particularly in commercial systems, is their capability of storing a complete
item of information on
one card, which may be handled separately or as part of a group, such as data on an insurance policy, a
BRL 1961, ANALYSIS AND TRENDS, start page 1032
|
payroll line, a stock item, a set of corresponding test data, etc. There is no doubt that punching cards is a slow process. Paper
tape perforators are also relatively slow in the sense that the data to be punched is usually available at a rate faster than paper
may be mechanically perforated, although high speed perforators are being developed and are finding application. Keyboard
input systems are useful primarily for the manual insertion of words for test or other special purposes.
In addition to paper tape and card readers and punches, many systems utilize high speed printers and magnetic tape units
as a medium of input and output. Magnetic tape output still requires a conversion from magnetic tape to cards or printed page
in order that the information be available to operating personnel. However, since human intervention is gradually being
reduced, the use of magnetic tape for input, output and storage is increasing rapidly. The prevalence of various input-output
media for the 222 computing systems described in this report may be determined by examining the data under the sub-heading
"INPUT" and "OUTPUT" in the systems descriptions given in Chapter II.
One method for decreasing machine time spent waiting for reading and writing instructions to be carried out is to
provide for concurrent operation. The later machines have built-in circuitry for permitting reading and writing to take place
during computations. Apparently the only stipulation is that a given storage location does not become involved in reading,
writing and computing at the same time. Many machines, for example, compute while punching and reading cards or while
"looking-up" information on tape. Others fetch the next instruction out of storage while performing an operation.
Another method of reducing reading and writing time and to avoid a large amount of lost time when a large amount of
machine reading and writing is necessary is to provide for reading and writing on a high speed device such as a magnetic tape or
wire unit and allow "conversion" to another medium to take place off the machine at "leisure". Magnetic tape-to-card
converters and inverters are becoming available as well as magnetic tape-to-printed-page converters. Paper tape and cards may
sometimes be considered as forms of storage, since information recorded on these media may be returned to the machine.
Considerable progress is being made in the field of printed page readers. See, for example, the IBM 1401 System.
It is often necessary to have computing systems capable of communicating with one another directly. For this reason,
input-output media conversion is becoming quite prevalent and large conversion equipment is rapidly becoming available. Input-
output schemes are so many and varied, that a complete treatment of this subject is beyond the scope of this report.
CIRCUIT ELEMENTS OF THE ENTIRE
SYSTEM
There are many impressions which come to mind when one examines such things as transistor, tube and crystal diode counts
in a large scale computing system. There is a tendency to visualize a large, sprawling system when the tube count is high. There
may be large tube-changing programs based on experience in effect on these large systems. Failure rates, preventive maintenance
techniques, tube life problems, design limitations and tube specifications must all be considered on a systematic basis when the
tube
count is high. Tube count and a knowledge of tube operating characteristics may yield an approximate estimation of some of the
problems that may be encountered in the operation of the system. Table X shows the approximate number of tubes utilized in
some of the computing systems described in this report. Maintenance of transistorized systems has become somewhat simpler
than maintenance of vacuum tube systems. Power and space requirements for transistorized systems are considerably reduced.
The servicing of a large electronic computing system can be materially simplified by reducing the number of tube types in
the system. Standards for tube testing need apply to fewer tube types and tube checking can be further systematized due to a
reduced number of test variations. Of course, a test specification or test criterion must be established for the most severe
application for which the particular tube type will be applied. A severe or special circuit requirement may be better served
through
BRL 1961, ANALYSIS AND TRENDS, start page 1033
|
the use of another tube type. This, then increases the number of tube types. Normally, it is, possible to select a type of tube for
a group of duties. In a given system, for example, a certain type is selected for driving, for voltage amplification, for flip-flop
circuits, normally "on" or "off" conditions, etc. This establishes a number of tube types for a given system and any
modification of the system usually should include this "tube type" complement.
The question of crystal diode reliability, diode testing techniques, and diode logical network design, such as individual clamps
versus wired plug-in units, become subjects of interest when diodes are utilized. The quantity of diodes in a given computing
system may be indicative of the nature of the servicing problem, but only when the failure rates, life and circuit demands placed
upon the diode are known. To some extent, malfunctions due to diodes can be aggravated by elevated temperatures. The printed
circuit logical package, containing a specific array of "And" and "Or" gates have become the most prevalent means of
fabrication. The extent of crystal diode use is shown in Table XI.
Many recently developed systems utilize transistors for driving, switching (gating) and other logical functions. Reduced
power and reduced space requirements are advantages of these systems. The question of reliability is rapidly being resolved, as
printed circuits and packaging techniques continue to be improved. Table XII shows the quantity of transistors utilized in the
various computing systems described in Chapter II.
CHECK FEATURES
The question of what type of checking features should be incorporated into a given general purpose computing system is
still being tossed about by various manufacturers. The type of built-in check varies from manufacturer to manufacturer and
from system to system.
It is usually possible to check all machine calculations by programming techniques. A well designed system can proceed for
many hours without a malfunction. If this is the case, it is entirely possible that the installation of a checking system can do
more
harm than good since the checking features can malfunction and cause an alarm or stoppage when a machine malfunction has not
occurred. For example, the second unit of twin arithmetic units can malfunction, the comparer of a redundancy checker can
malfunction, or a forbidden pulse combination decoder can malfunction, all. yielding false indications of a machine malfunction.
For those systems which do not have built-in checking circuits, the operator or programmer must program a check or the output
may be reviewed.
About 87,% of the 222 computing systems reported utilize some form of automatic built-in check. A redundancy or
duplication check is used in about 8% of the systems. Some type of overflow or exceed capacity is used on about 23% of the
systems and an odd-even parity check in one form or another is used on 50% of the systems. Interesting to note here is that in
1957 only 20% of the systems bad a form of parity check. Various kinds of transfer checks are used on 19% of the systems.
Approximately 28'% of the systems established a checking system by detecting pulse combinations which are not supposed to
occur anywhere in the system. Forbidden pulse combinations checking stations are scattered around the system, e.g. in memory
transfer points, recording stations, reading stations, etc. The various names that have been applied to this type of check are
forbidden pulse combination, unused order (instruction), unallowable order digit, improper operation code, improper command,
false code, forbidden digit, non-existent code, and unused code. There is a distinction to be made between the terms order,
instruction, and command. The preferred definitions are given in the glossary of computer engineering and programming
terminology, Chapter IV. The following table shows the approximate distribution of checking methods in the systems described
in this report. Many systems utilize more than one check technique.
BRL 1961, ANALYSIS AND TRENDS, start page 1034
|
Distribution of Automatic Checking Schemes Among 222 Different Computing Systems
Parity (arithmetic, transfer, storage, recording) 99
Overflow (underflow, exceed capacity, divide by zero, divide overflow, 47
oversized quotient)
Transfer (echo, compare, validity) 38
Non-existent command 19
Non-existent memory address 15
Redundancy (equipment, operations) 15
Character code (non-numeric, illegitimate char, "all ones", sign) 14
Forbidden pulse combination (general) 14
Arithmetic (Modulo 3, 4, 9, 25, residue) 13
Timing (clock, synchronism, jitter) 12
Count (hole, address, row, block, word, random error) 9
Non-existent device
Miscellaneous (instruction-data, logic, inactivity,
unwanted digit, free time) 7
No built-in check 26
Not reported 22
POWER, SPACE, WEIGHT, AND SITE PREPARATION
Important aspects of computing systems are the physical factors of power, space and weight.
Power requirements may very well dictate the physical location of a large computing system within a building, particularly when the
power
required is in excess of 50 Kw. For most systems, however, the power is brought to the most favorable computer location from the
point of view of
personnel accessability for operation and servicing. Table XIII shows the power requirement of various domestic digital computing
systems, operational or
about to become operational in the United States.
An interesting figure might be the relation between the number of tubes utilized in a computing system and the power requirement.
In order to
determine whether or not a consistent tube to power ratio could be established, the ratio was determined for the computing systems
for which the data was
available. For the vast majority of computing systems the tube-power ratio is approximately 710 tubes/kilowatt. A sample taken of
transistorized systems
shows that the ratio of transistor quantity to power is about 6,000 transistors/kilowatt.
The problem of space requirements has been solved in so many ways it is impossible to determine a consistent relation between space
requirement
and any other factor. Large computing complexes have been installed in areas ranging from a corner of a basement to an entire floor
of a large building. The
pictorial coverage of computing systems and the space requirements discussed under the sub-heading "POWER, SPACE, WEIGHT, AND SITE
PREPARATION" in the systems descriptions of Chapter II give the space requirements of the computing systems described in this
report. The dimensions of
various components of utilized systems are important when considering clearance in rooms, passages, doorways and elevators.
Air conditioning requirements vary considerably from system to system. Air conditioners for computing equipment may utilize water
to absorb the
heat from circulated air, use a secondary loop of air, force the heated air to the outside, or utilize an outdoor evaporator. The
smaller systems circulate room air
and depend on the ambient temperature to cool. Almost 100% of the power required by the system is dissipated in the form of heat
and must be removed. The
large systems usually require separate heat removal facilities. For many systems, humidity and dust control within the machine are
required in order to
maintain satisfactory operation.
The factor of weight can be important when the floor loading limits for distributed and concentrated loads are within the weight
range of the computing
equipment. Many systems may require reinforced or specially constructed buildings. Many items of peripheral equipment may cause
concentrated loads in
BRL 1961, ANALYSIS AND TRENDS, start page 1035
|
excess of maximum permissable concentrated loadings on some structures. Vibration and shock caused by some equipment such as
tabulators and card
punches can cause trouble in other components. Shock and vibration absorbing pads are required in such cases. When unitized
construction is used, the
weight of a single unit must also be considered when transporting and installing.
Many systems require extensive site preparations. Others may be "plugged in" to any convenient outlet. This topic is adequately
discussed in
the systems descriptions of Chapter II.
PRODUCTION RECORD
In almost any new and rapidly changing field there will be many instances in which an experimental prototype of a large piece of
equipment will be
built. This is the result of the normal course of events, namely, a feasibility study, a research effort, a development effort and
a prototype construction. Mass
production then occurs when the demand for systems is sufficient to warrant production in quantity.
A review of the sub-heading "PRODUCTION RECORD" will give an indication of the production status of various computing systems. The
quantity produced, the quantity in current production, in current operation, and on order are given. Delivery times quoted show
that immediate delivery
is now possible for many computing systems. Table II shows the quantities of the various systems that have been produced.
Information on unreported
systems was considered proprietary by the manufacturer.
COST, PRICE AND RENTAL RATE
Perhaps the most elusive and intricate item considered in the systems descriptions of this report is the question of initial cost,
blandly described as
"approximate cost of basic system". Manufacturers are quite naturally quoting current prices for their respective systems. The "one
of a kind" system usually
includes all research, development, construction, overhead and sub-contracting costs. The "basic system" usually includes minimal
input devices, the
controls, the storage system, the arithmetic unit, and minimal output devices. All conversion equipment such as card-to-printed
page (tabulators), card-to-
tape, tape-to-card etc. are considered peripheral equipment, and both the quantity and type is dependent upon specific system
application. These are not
included in the cost or price of the basic system. Prices of these may be found under "Additional Equipment". In order to determine
the cost of a given system,
refer to the system description. Table XIV shows the approximate relative cost of various computing systems. No attempt was made to
resolve or explain any
discrepancies between prices quoted by manufacturers and those quoted by users. It should be remembered that users prices reflect
old sales, rental rates
were established by contracts written years ago, manufacturers are offering discounts on older systems, charging greater service
rates for older systems,
offering educational discounts, etc.
The methods of computing system or component acquisition include direct purchase at a fixed price, direct purchase on a cost plus
fixed fee basis,
continuous rental, and rental with all or part of the rental applicable toward purchase. Most forms of rental include servicing.
Direct purchase can include a
service contract. Rental rates are of the order of 3 per cent of the direct purchase price per month. The sale and lease policy of
various manufacturers is given
under the sub-heading "COST, PRICE AND RENTAL RATE" in Chapter II.
Table XIV shows the nominal price one may expect to pay for a basic system. For many systems, one might add 20 to 80 per cent for
required
peripheral equipment. Most prices include installation but not shipping costs. Some of the figures reflect prices which are not
current and have not taken
into account general price rises during the past several years. Some figures include initial service or some type of warranty. The
figures quoted in Table
XIV are for general consideration only, and axe not for purposes of acquisition. Indeed, many systems are not available, even at
the price quoted, since the
price stated is actually the construction "cost" to the owner.
BRL 1961, ANALYSIS AND TRENDS, start page 1036
|
An attempt was made to discover whether a "system cost per tube" figure could be established. For the larger systems, the figure is
of the order of
200 dollars per tube installed and for the smaller systems approximately 100 dollars per tube. However, a glance at Tables X and
XIV will show that such a
figure can be calculated with some difficulty. An attempt to determine a figure such as "cost per cubic foot" of electronic
computing equipment would be
equally difficult. Such exercises are left to the reader should such figures be of any interest.
PERSONNEL REQUIRENTS
Personnel problems have confronted computing system operators and manufacturers from the very outset, in all phases of computer
research,
development, manufacture, installation, operation, improvement and servicing. Various grades of skills are required in the fields
of engineering, physics
and mathematics. Each large system has a crew of engineers and technicians for improving and servicing and a group of
mathematicians and operators for
problem analysis, coding and programming. In the very small systems, all of these functions may be performed by one or two persons.
The systems
descriptions in Chapter II show various estimates made by manufacturers and operators of what the personnel requirements are or
should be for various
systems. The estimates, in some cases, do not show the personnel required for overtime, vacations, illness and training purposes.
Just as in any application
of manpower to machines, it is necessary to provide sufficient manpower so as to maximize machine utilization whenever possible.
Many installations
consist of multimillion dollar computer complexes. Such large capital investments must be utilized at maximum efficiency in order
to avoid severe losses.
Twenty-four hour operation increases the daily output and provides for more efficient utilization of capital equipment. Ultimate
requirements for personnel
depend to a large extent upon the nature of the application, particularly as pertains to coders, programmers and analysts.
RELIABILITY. OPERATING EXPERIENCE AND TIME AVAILABILITY
The most discussed and most controversial issues in the field of computing machinery occur on the subjects of reliability,
efficiency and system
evaluation. The determination of the reliability of a system is difficult, primarily because of a lack of a common understanding or
interpretation of the
definitions of computer operating terms. What actually constitutes "good time" on a computing system? What is "down time",
"scheduled engineering",
"useful production and code checking"? An attempt has been made to provide working definitions of these and other terms in the
revised Glossary of
Computer Engineering and Programming Terminology given in Chapter V of this report. The very crude "Operating Ratio", as is used in
the systems
descriptions of Chapter II, is defined as the "Good Time" obtained on the machine divided by the total time one actually "Attempted
(or Wanted) to Run"
the system. The question arises as to where to put the time lost in scheduled engineering (preventive maintenance), since
technically, one is not
attempting to run the system during this period, yet the system is not actually "down". Many systems, are operated for 168 hours
per week. The operating
ratio for these systems would require that 168 be used as the denominator and the number of useful output hours as the numerator,
yielding a much smaller
(but perhaps truer) ratio than a system operated on an 8-hour 5-day week shift and using off-time for servicing. This latter type
of operation may yield
operating ratios of the order of .90 to 1.0 and give a false indication of reliability.
The question of how one determines the average error-free running period is also a difficult one. It may be estimated or calculated
by actual counts of
the periods of malfunction-free operation. It may be the period used as a guide by coders to prevent losses due to running for
extended periods between
obtaining output information, particularly where volatile storage media are being used. Mary questions regarding the subject of
"RELIABILITY,
OPERATING EXPERIENCE AND TIME AVAILABILITY" are answered under this subheading in the computing systems descriptions given in
Chapter II. A search of the system descriptions under this subheading will reveal those installations which have computer time
available to organizations
outside of the operating organization.
BRL 1961, ANALYSIS AND TRENDS, start page 1037
|
Many computing systems are approaching the age of retirement and replacement. Constant improvements have already
replaced many of the original components of a system. The next few years will see the retirement of many of the older
systems. Such retirement may take the form of salvage of parts, use for educational and training purposes, or scrap. Many older
models are available at reduced prices. A used computer market is developing. In accepting a used computer, one must be
prepared to accept a few headaches. Table XV shows how long some models of computing systems have been in existence.
ADDITIONAL FEATURES AND REMARKS
Under this subheading has been placed general information concerning specific computing systems
which did not have a "place" in the previous fourteen subheadings. Included under this subheading are
remarks concerning the pictures, information which arrived too late to be added to the system descrip-
tion under a proper heading, special features of the system and other miscellaneous items of information.
Under this subheading one will find what manufacturers and users considered to be the outstanding
features and unique system advantages of the particular system. Under this subheading are remarks
concerning the labelling,storage, shipping and protection from humidity, temperature and physical,
electrical, fire or other damage of magnetic tapes.
FUTURE PLANS
The electronic digital computer field is a dynamic one. Plans for acquisition and improvement of systems and
components are continually being made and modified. The plans of various operators and manufacturers are given under the
subheading "FUTURE PLANS" in the systems descriptions of Chapter II. Interesting to note are the transitions to new
systems being made by many users. "Second generation" (solid state) computers are now at hand.
INSTALLATIONS
A primary source of information concerning electronic digital computing systems is the operating organizations. The
acquisitional and operational problems of one organization may have already been solved in one way or another by other
organizations. Benefiting from the experience of others can be profitable, if only to avoid mistakes. Under the subheading
"INSTALLATIONS" in the systems descriptions of Chapter II, a list of the owners and operators of specific systems is given in order
that contacts between owners and. prospective owners may be established. Many co-operative "plans" have come into existence,
under which owners or operators of specific systems have engaged in sharing computer experience. Many computer sharing
contracts have been drawn and many computer centers have been established, offering computer time and personnel for the
solution of customers' problems.
--------- end table -------
|