distributed computing systems: an overview
TRANSCRIPT
University of Central Florida University of Central Florida
STARS STARS
Retrospective Theses and Dissertations
1977
Distributed Computing Systems: an Overview Distributed Computing Systems: an Overview
Haim Schwarzkopf University of Central Florida
Part of the Engineering Commons
Find similar works at: https://stars.library.ucf.edu/rtd
University of Central Florida Libraries http://library.ucf.edu
This Masters Thesis (Open Access) is brought to you for free and open access by STARS. It has been accepted for
inclusion in Retrospective Theses and Dissertations by an authorized administrator of STARS. For more information,
please contact [email protected].
STARS Citation STARS Citation Schwarzkopf, Haim, "Distributed Computing Systems: an Overview" (1977). Retrospective Theses and Dissertations. 376. https://stars.library.ucf.edu/rtd/376
DISTRIBUTED COMPUTING SYSTEMS: AN OVERVIEW -.-
BY
HAIM (JIMMY) SCHWARZKOPF B.S.E., Florida Technological University, 1976
RESEARCH REPORT
Submitted ih partial fulfillment for the requirements for the degree of Master of Science in Engineering
in the Graduate Studies Program of the College of Engineering, Florida Technological University
Orlando) Florida 1977
DEDICATED TO MY PARENTS
Dr. and Mrs. Bedrich Schwarzkopf
who made it all possible
DISTRIBUTED COMPUTING SYSTEMS: AN OVERVIEW
ABSTRACT
by
Haim (Jimmy) Schwarzkopf
Associative processors, paraJlel processors, content
addressable parallel processors, networks, and other architectures
have been around the computing scene as "Distributed Processing",
for some time now. Several hundred papers have been written
discussing their use and design but so far no academic work has
tried to summarize the field called "Distributed Processing .. using
a systems approach.
This research report attempts to remedy this lack. It
attempts to gather into one place information that existed as of
late 1976 in a format easily understandable by managers and systems
engineers. The report deals also with certain issues of central
ization and decentralization of EDP (Electronic Data Processing)
facilities, created by the introduction of distributed computing
systems into industries and businesses.
iii
ACKNOWLEDGEMENTS
I would like to thank the members of my committee, Dr. C. S.
Bauer, Dr. H. I. Klee, and Dr. B. W. Lin, for their kind attention
and scholarly advice. I am especially indebted to nr. Christian
Bauer for the many hours spent advising me on the development and
presentati-on of this report. Not only did he give freely of his
time in helping me on this report, but he was responsible for getting
me interested in computers in particular, and systems science in
general.
Finally, I would like to thank the entire C.O.E. faculty for
the excellent program of studies that has been offered to me.
Special thanks to Dr. George F. Schrader, Chairman of the IEMS
Department, who put up with me for the last four years and provided
teaching assistantships which helped to support me during this
period.
TABLE OF CONTENTS
ACKNOWLEDGEMENTS
LIST OF TABLES
LIST OF FIGURES
. . . . .
1 .
2.
3.
4.
5.
6.
7.
8.
9.
INTRODUCTION . . .
DISTRIBUTED SYSTEMS
HORIZONTALLY DISTRIBUTED COMPUTERS
(a) Parallel Processor .... (b) Associative Processors
VON-NEWMANN MACHINE . . . ..
VERTICALLY DISTRIBUTED COMPUTERS
BASIC NETYJORK TYPES
(a) Point-to-Point (b) Multipoint . (c) Centralized or Star . (d) Tree ...... . (e) Loop or Ring .. . (f) Mu1 ti star . . . .
DISTRIBUTED PROCESSING .
HARDWARE COMPONENTS
DISTRIBUTIVE DATA BASES
iv
iii
vi
vii
1
3
6
6 9
10
12
14
14 14 15 16 17 18
19
21
27
10. A REPRESENTATIVE DISTRIBUTED PROCESSING APPLICATION. 31
11 .
TABLE OF CONTENTS - Continued
CENTRALIZATION vs. DECENTRALIZATION OF COMPUTER SYSTEMS
12. CONCLUSIONS .
BIBLIOGRAPHY . .
v
36
46
48
TABLE
1
2
3
4
5
6
7
8
LIST OF TABLES
General and Organizational Considerations; Advantages . . . . . . . . . . . . . . .
General and Organizational Considerations; Disadvantages ....
Cost Factors; Advantages
Cost Factors; Disadvantages
Personnel Considerations; Advantages
vi
Technical Considerations; Programming; Advantages ..
Technical Considerations; Disadvantages ....
Technical Considerations:
Programming;
Operations
38
39
40
41
42
43
44
45
1 •
2.
3.
4.
5.
6.
7.
8.
9.
10.
11 .
LIST OF FIGURES
Division by Types of Distributed Processors
Array Computers
Multiprocessor Computer
Multicontrol Computer
von-Neumann Machine . . . . . . . . .
Point to Point Network ..... .
Multipoint Network ..
Centralized or STAR Network
Tree or Hierarchical Network .
Ring .or Loop Network ... .
MultiStar Network ... .
12. Distributive Processing Environment
v.i i
5
7
8
• • • • 9
. . 10
. 14
. 15
16
. . 16
17
. 18
• 35
1. INTRODUCTION
The field of Distributed Processing is eliciting a great deal
of interest at the current time. The task undertaken by this
research report .on Distributed Processing was one of searching and
culling the literature on three major topics:
* Distributive Systems and Computer Networks
* Distributive Data Bases
* Centralization vs. Decentralization of
computing systems
Distributed systems hold a special place among currently
feasible computer configurations, simply because they present a
new and attractive alternative to totally centralized or decentralized
systems. Unfortunately, the term 11 distributed processing" means
different things to different people. In the first sections of the
paper a brief explanation of the different hardware configurations
are given.
Distributed data bases are a very important part of the success
ful Management Information System. They are created by taking
portions or subsets of the overall corporate data base, and putting
them out in the remote locations. The remote site corresponds to
where the data is created and used, and where decisions are made
based upon it. Exception and summary data, which the headquarters
location needs for its data base can be retrieved from the remote
2
sites at appropriate time intervals. A part .of the paper describes
and explains Distributed Data Bases and its uses.
The organizational authority structure is very important in
determining the chosen information system configuration. Those
organizations accustomed to central control move earliest and most
strongly to centralized data processing; those most devoted to
decentralization move slowly, carefully and with maximum compromise.
The last section of the report covers the EDP related issues in
centralization or decentralization of companies affecting Distributed
Processing.
3
2. DISTRIBUTED SYSTEMS
During the research upon which this report is based, it was
found that highly parallel processors, computers-on-a-chip, networks,
intelligent terminals and others have all been described as
11 distributed" systems. For the purposes of discussion and ease of
understanding, the systems covered by the term distributive proces
sing will be divided in two main groups: (a) horizontally distri
buted systems, and (b) vertically distributed systems.
Before defining the two main groups of distributed machines,
the difference between a real DISTRIBUTIVE SYSTEM and one that is
only DISTRIBUTED should be ·noted.
The phrase 11 DISTRIBUTED PROCESSING 11 stands for the use of a
DISTRIBUTIVE SYSTEM, which entails the segmenting of its data bases
and distributing its processing among smaller modules. This system
can be either distributed geographically or the data base segments
and the processor modules could be clustered in only one location.
There are many ways of accomplishing the distribution of
processing. The easiest and most clear division is in the amount of
processing that takes place simultaneously on the "same 11 application
or program. This concept was used when dividing distributed pro
cessing in two main groups.
Horizontally distributed systems are those systems which can
process data blocks of the same application simultaneously, while
4
vertically distributed systems process different parts of the data,
at different levels, sending results for further computation from
one to anoth~r_ of the processors.
Figure 1 depicts how these processing concepts have been
classified for the purposes of this report.
Poi
nt
to
Poi
nt VE
RTIC
ALLY
DI
STRI
BUTE
D PR
OCES
SING
Mul
tipo
int Sta
r or
C
entr
aliz
ed
I A
rray
DIST
RIBU
TED
PROC
ESSI
NG
Hi e
rarc
hi ce
il or
tre
e Lo
op
or
Rin
g
Mul
ti st
ar
HORI
ZONT
ALLY
DI
STRI
BUTE
D PR
OCES
SING
Non
-Ass
ocia
tive
Par
alle
l M
ul;i
proc
esso
r M
uiti
co
ntro
l
I P
aral
lel
Fig.
1.
D
ivis
ion
by
Type
s of
Dis
trib
uted
Pro
cess
ors
Ass
ocia
tive
I
I N
on-
Para
11 e
1
6
3. HORIZONTALLY DISTRIBUTED COMPUTERS
This family of computers are the newest and the most mis
understood architectures. As is seen in Figure 1, it is made up
mainly of parallel and associative processors with some overlapping.
{a) Parallel Processors
A very rough definition of parallelism is that of putting
together N computers to form a supercomputer. This can be done in
terms of either parallelism within the instruction stream or
parallelism within the data stream or both. This combinations of
parallelism produces 3 types of systems: multicontrol, array,
multiprocessor systems, respectively (Kuck, 1977).
Array computers: this type of a system operates on vectors
as basic units of information. Its parallelism derives from a
parallel data-stream with a single instruction stream. The pro
cessing power is distributed in that while it executes a single
instruction stream {with loads, adds, stores and branches of a
serial von-Neumann computer), it manipulates whole vectors of data
simultaneously, as shown in Figure 2.
An example of this kind of computer is the ILLIAC IV computer
{Thurber and Wald, 1975).
Fig. 2. Array Computers
SOURCE: Reddi and Feustel, 1976
7
AM
AM
AM
n 0 3:3: 3:rrl C3: :z 0 ...... ::c n-< )>I -t-t t-40 0 I :z -o V>;o
0 n n 0 rT1 z V) -tV> ;co 0 ;o r-
t ~~
MEMORY
Multiprocessor Computer: this type of a system consists of
N complete computers plus interconnections for passing data and
control information among the computers. It derives its parallelism
from having both a parallel data stream and a parallel instruction
stream. This computer is shown in Figure 3.
An example of this type of system is a machine developed
at Carnegie Mellon University (Enslow, 1977).
CONTROL UNIT 1
..
CONTROL UNI.T ;-2
CONTROL UNIT ____ ......,. N
PROCESSOR
1
ROCESSO 2
ROCESSOR N
(
8
DATA STREAM 1
DATA STREAM 2
DATA STREAM
N
Fig. 3. Multiprocessor Computer
SOURCE: Reddi and Feustel, 1976
Multicontrol Computers: In this type of computers each
operand is operated upon simultaneously by several instructions. It
derives its parallelism from having a multiple instruction stream
but a single data stream. This computer is shown in Figure 4.
An example of this type of system is the CDC STAR computer
(Bell, 1971).
9
CONTROL UNIT Instruction 1 1 >
PROCESSOR ~ 1 .
. CONTROL Instruction
UNIT 2 > PROCESSOR '~
2 -l.
CONTROL UNIT n
Instruction n
) ~-P-R_o~_E_s_so_R _ __,~
Fig. 4. Multicontrol Computers
SOURCE: Reddi and Feustel 1976
(b) Associative Processors
DATA STREAM
A very rough definition of ASSOCIATIVE PROCESSORS, also called
Content Addressable Computers, is any machine in which the processing
units (or processing memory) are addressed by a property of the data
contents instead of the memory address itself (Yau and Fung, 1977).
This type of computer will be important in the next generation of
business machines (Foster, 1976). Associative processors can be
either parallel or serial. Examples of associative-parallel computers
include the ILLIAN IV and the Goodyear STARAN machines (Higbie, 1976).
10
4. THE VON-NEUMANN MACHINE ~.-
In 1946, John von Neumann, a mathematician from the Institute
for Advanced Study at Princeton University described a basic
philosophy of computer design (Stone, 1975). Almost everything
concerning computer design that von-Neumann discussed in his paper
has been incorporated in modern computers. Thus, it is often stated
that the basis for the design of computers is the "von Neumann
Machine".
INPUT ) PROCESSOR OUTPUT
1 f Data/Instruction ,-------~-- Flow
STORAGE
Fig. 5. Von-Neumann Machine
SOURCE: Stone, 1975
This machine works on the basis of a single data/single
instruction stream sequential organization (Bell, 1971). The
processor has a control unit that identifies input information
either as data or instruction, both of which have to be stored in a
sequential order. These today are the most common machines and are
11
the building block for the vertically distributed computers.
Vertically distributed computers are the real force behind dis
tributed proces~ing (March, 1976), that is why the rest of the
research focused nearly in its entirety in this kind of system.
From this point on, in the paper, the terms Distributed Processing
and Vertically Distributed Processing will be used interchangibly.
12
5. VERTICALLY DISTRIBUTED COMPUTERS
This family of distributive computers, better known as
computer networks can be defined as an interconnected group of
computers, where each either acts as a processing system or as a
communications control system. Computers are generally of the single
data stream/single instruction stream.
The computers in operational control of the network can be
divided in two groups (Nielsen, 1974):
(a) main-site, or host computers
(b) remote computing systems
The host processors in the network perform major computation, control
data bases, and generally supervise operation of the network. They
can share such resources as programs, data bases, and memory space.
Remote computing systems are systems with access to the host
processor, that perform a minor part of the processing before sending
the 11 edited 11 data to the host processor.
The communications control computers are devoted exclusively
to network control functions. These functions include line control,
error checking, message formatting, message switching, and data
con centra ti on.
In addition to the processing and control computers, a typical
network might consist of a wide variety of remote terminals, each
with some processing capability.
13
To understand vertically distributed processing systems,
first some common network terminology should be defined (Kimbleton
and Sneider, l975):
TOPOLOGY = refers to the geometric arrangement of
links and nodes of a network.
LINK = is the communications path between two nodes
NODE = is the end point of any branch of a network.
In designing a network, many factors must be evaluated in
choosing the most suitable topology. However, one major factor
exerts a pronounced influence on this choice: the type of partic
ipation by each of the nodes. Any node can be a provider of
resources exclusively, a user of resources exclusively, or some
combination of resource provider and resource user (Lynch, 1976).
14
6. BASIC NETWORK TYPES
(a) Point to Point: This is the simplest network (Figure 6) made
up of a host processor connected to one communications input/output
device per line. The communication input/output device may be a
terminal or another processor {Nutt, 1975).
HOST COMMUNICATIONS AND/OR PROCESSOR I/ 0 DEVICE
Fig. 6. Point to Point Network
SOURCE: Moore, 1977
(b) Multipoint: In multipoint operation, one station in the network
{normally the host processor) is always designed as the control
station (Figure 7) {Nutt, 1975).
The remaining stations are designated as tributary stations.
The control station controls network traffic by means of polling;
that is, it polls the tributary stations · {which may be terminals
or computers) to send messages. Messages can either go only between
the control station and tributary stations or between all stations.
~
HOST PROCESSOR
*TRIBUTARY STATIONS
Fig. 7. Multipoint Network
SOURCE: Moore, 1977
15
(c) Centralized or STAR: In this type of system all users comrnun-
icate with a central point that may have supervisory control over
the system. Data movement is outward or inward toward the host
(Figure 8). If communication becomes necessary between the remote
processors or terminals, the host acts as a central message switcher
to pass data between them (DEC, 1974).
HOST PROCESSOR
Fig. 8. STAR Network
SOURCE: Moore, 1977
*TRIBUTARY STATIONS
16
(Remote Processors)
(d) Tree: When unlike components are connected in a vertical
(Hierarchical) distribution of functions, especially control, a
system like that of Figure 9 res~lts (DEC, 1974).
HOST PROCESSOR
*TRIBUTARY OR REMOTE STATIONS
Fig. 9. Tree or Hierarchical Nen~ork
SOURCE: Moore, 1977
17
(e) Loop or Ring: In this arrangement, the communication bus is in
a ring configuration shared by all stations. The advantage of this
architecture lies-in the high reliability of the bus, as for example,
technical difficulties in any one point in the ring will not cause
total communications failure. The system is shown in Figure 10
(Acree, 1976).
HOST PROCESSOR
Fig. 10. Ring or Loop Network ,
SOURCE: Moore, 1977
*TRIBUTARY OR REMOTE STATIONS
18
(f) Multistar: In this configuration there are several supervisory ·
or exchange points, each with its own set of host and remote pro-''·
cessors and a means for direct communication between the points
(Figure 12) (DEC, 1974).
~~*
HOST PROCESSOR
I
HOST PROCESSOR
*
HOST PROCESSOR
*TRIBUTARY OR REMOTE TERMINALS
HOST PROCESSOR
Fig. 11. Mu1tistar Network
SOURCE: Moore, 1977
19
7. DISTRIBUTED PROCESSING - . -
For the great majority of computer professionals, the different
processing architectures described in the proceeding sections are
only concepts whose importance lies in the understanding of possibil
itie~ in the future of data processing.
Vertically distributed processing has caught on very fast and
with the proliferation of this type of system, it has grown to the
point where literature, applications and developments are ignoring
the horizontally distributed processors. For the reason mentioned
above, the words distributive processing will be taken to mean
vertically distributed processing in the rest of the report.
Distributed computing is characterized by two distinct but
(Black, 1976) closely related forms of processing; communications
processing and dispersed data processing. Communications processing
provides an intelligent pipeline that permits the effective transfer
of data and control of the machine/machine interfaces. Dispersed
data processing supports the man/machine interfaces that interact
directly with the user.
Both are important, but most of the emphasis and support to
date has been for the communications function because it must be
well (Kimbleton and Sneider, 1975) developed and integrated for a
distributed processing network to work even reasonably well. For
this reason, network definitions and implementations are undergoing
20
rapid evolutionary development.
A network structure is chosen simply to support the goals
and functions that an organization wishes to implement; it is not an -.-
end result in itself. Most network organizations are generally
reliable with existing hardware and software. They are like social
organizations in that they are organized either hierarchically, with , a powerful central computer acting as a feudal overload of the system,
or anarchically, with the system composed of an association of
independent computers.
It is important to note that all kinds of networks can be
constructed using the same heterogeneous components, although some
manufacturer's network philosophies are predisposed to a specific
structure (the multistar structure may require "some 11 not standard
off-the-shelf hardware and software products) (Hovey, 1976).
21
8. HARDWARE COMPONENTS
When analyzing the hardware components peculiar to distributive
processing as opposed to more traditional approaches, two types of
components have to be examined (Cooper, 1977).
The first category consists of the processing elements being
linked together to form the network and where defined earlier as
Dispersed Data Processing Equipment. When surveying this category,
the question to keep in mind is: 11 Are there particular types of
equipment more apt to be linked together in distributed networks
as opposed to centra 1 i zed networks? .. (Lynch, 1976)
The second category consists of the communication link itself,
previously defined as Communications Processing Equipment. Focusing
on the equestion of whether there is communication hardware peculiar
to distributed networks because of the distinctive characteristics
of this type of networking. Devices common to all networks are:
(a) Dispersed Data Processing equipment:
Te 1 etypewri ters
CRTs or video display units
Hand-held terminals
Remote batch terminals (high-speed input and output devices)
Intelligent terminals (CRTs equipped with cassettes, stored program capaci ty and even disk storage)
22
Workstations (clustered terminals ·around a terminal controller)
Industry-application-oriented terminals ---Graphics terminals and plotters
Office computers (designed for magnetic ledger operations)
Small business computers (often minicomputer-based)
Mini computers
General-purpose computers
(b) Communications Processing Equipment:
Corrmon carriers
Multiplexers
Concentrators (often part of clustered terminal packages)
Private branch exchanges (for line switching)
Message switchers
Communications controllers
Communications processors
Communications software
Most distributed processing systems will be built using
minicomputers and small business computers (Kallis, 1977). The
philosophy behind the use of minis - give a user only what he needs
to do a particular job - is similar to the philosophy behind
distributed processing - make the data processing facilities fit .
the applications.
Two newer types of business minis, the personal computers and
23
the word processors, particularly emphasize this trend toward local
processing and the distributed networks. Both are designed for
nons peci a 1 i zed·, -ncrnprogrammi n g office personne 1 , and both cons e
quently present interesting management and control problems (Burns,
1977).
Although many factors have combined to make distributed
processing a reality, none has contributed more than minicomputers.
The minicomputer proved to users that dispersing computer power
resulted in quicker turnaround and lower costs. Experience has I
shown that while super hosts were burying user programs in long
batch processing job queues, the minicomputer users were getting
responses in a matter of seconds, usually interactively.
Bureaucracy was not only eliminated at the computer site,
but also during the purchase of the system. The cost of a typical
minicomputer system ($8000 to $150,000) is low enough that the
system can be considered a tool rather than a major capital
expenditure. Thus, fewer management levels need to be consulted
when making a buy decision.
I
If minicomputers and intelligent terminals of all sorts are
particularly characteristic of distributed systems elements, com
munications processors are particularly characteristic of network
management hardware. There are three main types: front ends, con-
centrators and message switchers.
A front-end processor, by definition, is located with and
24
attached to one or two specific host computers. ·Its primary function
is to conserve the host computer's resources by assuming some or all
of its communications management responsibilities. As a result, -.-
data received from a network can be presented to a host in a constant
format and from a single defined source. The host can be essentially
free of any direct involvement with network requirements.
In a distributed network, the front-end processor is used in
much the same capacity, but it can interface its host computer(s)
with the network's message-switching element instead of directly
with terminals. For this function, the front-end processor intercepts
and controls all traffic between its host(s) and the network. In
addition, the front-end processor can act as a controller for terminals
and peripherals that are direct subordinates of the host(s).
The remote concentrator, or line concentrator, can be linked
to a front-end processor located away from its host computer rather
than adjacent to it. In fact, many mini-based communication processing
devices can be configured either as front-end processors or as remote
concentrators.
In a distributed network, the remote concentrator can be used
either to offload the host computer and to improve line utilization
or as a network node. On the other hand, a general-purpose mini
computer or small business computer used as a network node can be
configured to perform remote concentrator functions.
In networks where a large number of terminals exist at the
remote site, the remote concentrator may be required to perform polling
25
functions, in which case it would be involved in .two different levels
of polling: polling terminals and being polled by the host computer.
Message swttchers have been variously called traffic directors,
message routers and dispatchers. Unlike that of the remote concen
trator, the message switcher•s output is not necessarily to a host
computer, but to a terminal, remote concentrator or another message
switcher.
A message switcher can be based on either a general-purpose
minicomputer or a special-purpose processor.
Circuit switching by the host computer in a distributed
environment is strictly limited and is usually alone only as an
emergency backup. In networks where a number of remote terminals
have to communicate with one another, as well as with the host com
puter, the use of a message switcher is cost-effective.
One major feature of a message switcher which a remote con
centrator generally lacks is the ability to detect and act on a
priority indicator. Also, the message switcher frequently acts on
administrative data and text messages, whereas the remote concentrator
typically acts on data intended to be processed in a computer.
Most users do not buy a whole distributed system. Instead,
they take an existing network, perha~s add more intelligence to
certain elements, change the software in the host and make other
changes necessary to disperse processing. The impetus for these
changes probably came from a need to expand. Instead of upgrading
the host, a remote mini is added and the network architecture is
26
altered.
Selection of distributed processing system components should
procee~ from performing a study to determine functions and tasks
appropriate or desirable for distribution, to a consideration of
what classes of devices can perform those tasks and functions, to an
evaluation of the various manufacturer's product offerings.
The operational organization of the business is the key to
determining if distributed processing is suitable. The business
structure also holds the answer to component selection once
distributed processing has been decided upon (Doll, 1977}.
27
9. DISTRIBUTED DATA BASE ---
The tenm distributed data base can be taken to mean either
distributing the data base management function (the control and
manipulation of data) or distributing the content of a data base
(the data itself). These are two very different techniques with two
very ifferent realizations. The data base management function
throughout a network is still in the initial implementation stage
(Champine, 1977). Today the majority of disdtributed data bases
simply mean redundant data. Local copies of data are maintained
at user sites, with most of the same data still retained in a
centralized data base.
Most centralized data bases operate under the integrated
corporate data base (ICDB) concept. Before considering the dis
tributed data base, it might be well to review the elements of a
centralized data base, the ICDB.
An ICDB environment can be defined as the consideration of the
collection, storage and dissemination of data as a logical, centrally
controlled and standardized utility function (Curtis, 1977).
It should be emphasized that ICDB is not a system; rather it
is a concept under which the information system structure should be
implemented. This implementation, using the ICDB concept, requires
development of the four functional elements (Yasaki, 1977). The four
elements of the ICDB are:
28
1. Data Bank- the logically centralized repository
all the data utilized in a corporation.
2 . . _Data Base Administrator (DBA) - a person responsible
for coordination of all data-related activities.
3. Data Base Management System (DBMS) - a software
function performing the storage, retrieval and maintenance of data.
4. User/System Interface (USI) - the subsystems necessary
to permit multiple classes and types of users to direct the system to
structure the available data effectively into information and thus
communicate with and fully utilize the resources at their disposal.
Since a centralized data base is a collection of logically
related files at one location, then a distributed data base is a
logical integration of related data bases at a number of locations.
Tnere are various reasons for the necessity of . this integration, two
of which are paramount: it permits users to produce reports
summarizing information from different locations and it provides a
means of employing data stored in another data base. Neither of
these reasons precludes a distributed data base with no redundant
data. And although such a setup is technically feasible, it is
operationally undesirable for reasons of backup, security and
integrity (Korns, 1976).
' There are some design considerations for a sound, shared,
distributed data base. First, the design must be compatible with all
systems within the network. The data base organization~ to be easily
implemented and maintained, should be standardized, and the accessing
29
languages and data access methods should be compatible with all systems.
To accomplish this, some major problems will have to be solved.
For example, the- DBA must specify the correct data content and logical
organization for each user node; a method must be devised to direct
all users to all data located around the network; and safeguards must
be developed to ensure that all those requesting data are qualified
(Rodriguez, 1976).
The integrated corporate data base concept could be applied to
support an organizationally distributed data base, as well as a
centralized one. Furthermore, DP installations with a multivendor
hardware policy may also take full advantage of the ICDB concept by
logically centralizing the data bases of different vendors• processors
into a single data base.
Storage facilities housing the data base should be designed to
support a full range of storage and accessing requirements. This
can be facilitated by using hierarchial secondary storage.
Different types of secondary storage media offering alternative
access techniques and speeds at correspondingly adjusted costs would
allow the DBA to specify and design the optimum physical storage
configuration for the data base. By providing multiple levels of
hierarchy, an installation can capitalize on the unique requirements
of individual users and save both time and money in the process of
storing and transferring data.
Total system efficiency depends largely on the specific
organization of the data base. There can be only one physical
30
representation of the data (random or sequential), and the DBA•s
choice as to which representation to employ is important if not
critical. ---
The applications view of the data (sometimes referred to as
the logical representation) is equally important, inasmuch as the
application modules of the user systems will be designed to utilize
these representations. However, the amount of flexibility allowed
in logical representation for applications depends almost entirely
on the specific implementation of the DBMS (Tsichritizis and Cockouski,
1976).
31
10. A REPRESENTATIVE DISTRIBUTED SYSTEM
A laboratory computer complex can be used to illustrate some
of the principles mentioned in the previous sections (Karp, 1976).
In this example, the computers are attached to analytical instruments,
used to control laboratory experiments and perform data acquisition.
One small computer is dedicated to a nuclear magnetic residence
spectometer, another to an infrared spectometer, and a third to a
mass spectometer. An application program assigned to the computer
system in each instrument manages the experimental apparatus.
The differences between a set-up of stand-alone computers
that partition a data processing workload and a computer network that
integrates these same tasks will now be shown.
In a stand-alone system each instrument's readings are
independent of each other, and either a manual set-up or another
independent computer system would be necessary to merge the analytical
data to determine, let us say, the molecular structure of a complex
organic compound.
Tie the computers together into a network, however, and they
can then do the analysis online. Moreover, the central computer by
maintaining data files on all experimental results can prevent a
researcher from inadvertently repeating an experiment. But even
more important, the research team would have instant access to all
the cumulative data and analyses on a sample because this information
32
could all reside in a comprehensive network file.
In this laboratory example, the system designer also has an
option to use one- ef two methods to transmit the data around the
system. The dedicated computers could store all the analytical
data generated in an experiment and then use high-speed synchronous
communications to transmit the information to the central computer.
Alternatively, the data could be transmitted piecemeal as determined
by the program using a synchronous data communications techniques.
Either data transmission method requires that the computers
be compatible, and this constraint is typically achieved by means of
software designed to follow a so-called communications protocol. In
other words, all processors linked in a network must 11 Understand 11
all transmitted and received messages. Such compatibility should be
beneficial to users. Networks that are constructed from a diversity
of computer brands tend to require more than one protocol, and this
can cause transmissions delay. Before re-transmitting a communication,
a processor will have to reformat the received message so that it can
match the protocol required by the non-compatible computer.
Networks can perform other functions besides distributing a
data processing workload. Many networks are used to provide redundancy
and back-up as a security measure. And even when setting up stand
alone computers, designers should consider creating a system that
could be integrated into a network to meet unforeseen or expansion
needs. The designer can build a network in increments according to
need and cash availability.
33
As was explained a network distributes computer functions
among its elements according to the organizational structure, locations,
and tries to get -tfle most cost-effective arrangement in the specific
application (Benson, 1976).
An example of the above statement can be seen in the mock-up
of a distributed processing network within a single company depicted
in Figure 12. This company is made up of Research & Development,
Manufacturing and Sales (Karp, 1976).
Scientists in the research and development division work on
laboratory experiments and gather data for analysis. Each lab uses
a dedicated minicomputer having disk storage and a printer; but the
cost of individual computers are high. So, a resource sharing
computer network is employed. The large, expensive peripherals reside
at a central host computer. Each lab has a relatively small satelite
system that includes a console terminal. Scientists use the terminal
to edit, compile and transmit data and programs for storage and
execution.
The manufacturing division, aiming to automate plant operations,
wants to control raw material input, the manufacturing machinery,
and also operation of an automated warehouse for finished goods
storage. The management also wants access to current inventories
and stock levels. A hierarchical distributed computer network can
distribute the tasks to individual computers, each specifically
designed to handle a function and each in communications with all
other systems. For example, realtime process control systems control
34
the actual manufacturing operations, and are in turn linked to
supervisory systems that control overall parts flow. The supervisory
system computers .pass data to a transaction-processing or other large
system for management control.
The sales offices that are widely dispersed need data to access
current inventory information and shipping dates, not individual
telephone lines to a central computer are expensive. A communications
network that significantly reduces line costs is installed. It
contains terminal concentrators at local regional centers to switch
data traffic over high speed lines to the central system. These
data can be terminal input or data-pre-input on disk or magnetic tape.
Processed information can be shipped back to the offices, to terminals
or disk storage for later printing offline.
In the example, the mechanism for data communications between
programs and devices on different systems and on computers running
under different operating systems would be the relatively high levels
of software sophistication and expense.
RESE
ARCH
&
DEVE
LOPM
ENT
Smal
l Sm
all
Cam
p.
Com
p. I
Smal
l M
ini
(Hos
t)
Com
p. Co
mp.
Loc
atio
n 1
The
com
pute
rs
here
sh
ould
be
HORI
ZONT
ALLY
DI
STRI
BUTE
D fo
r fa
ster
re
spon
se
tim
es.
Min
i
/(H
ost)
Co
mp.
Smal
l Co
mp. '
Ter
min
als
Smal
l Co
mp.
Wes
tern
R
egio
n S
ales
I T
erm
inal
s
Rin
g or
STA
R S
truc
ture
d N
etw
ork
MAN
UFAC
TURI
NG
Mic
ro
Mic
ro·
Mic
ro
Cor
p.
ern~ C
omp.
M .
. ( H
t)
Sm
a 11
1n1
os
Comp
Hie
rarc
hica
l S
truc
ture
d N
etw
ork
Com
p. ·
~Micro
Com
p.
Loc
atio
n 2
CENT
RAL
EDP
Mul
tist
ar S
truc
ture
d (H
OST)
N
etw
ork
INST
ALLA
TION
~
·
~
Eas
tern
R
egio
n S
ales
Smal
l Co
mp.
I T
erm
inal
s
Fig.
12
. D
istr
ibut
ed P
roce
ssin
g.
Min
i /(
Ho
st)
/ Co
mp.
Smal
l Co
mp.
I T
erm
inal
s
""C
:z
0
(l)
....
..
~~
0 -s<
.n
" c-
t- -s
s::::
n e-t c:
-s
ro
0..
SOUR
CE:
Kar
p,
1976
E
nviro
nmen
t: w
ith H
oriz
onta
l &
Ver
tica
l C
ompu
ters
w
01
11. CENTRALIZATION AND DECENTRALIZATION OF COMPUTING SYSTEMS
36
The issue of centralization of information systems has been
unleashed by the increasing number of Distributed Computing Systems.
Furthermore, the centralization-decentralization problem is complex
and important enough, so that any paper dealing with Distributed
Processing should cover at least the most important aspect of the
problem: the centralization-decentralization decision.
Articles on the advantages and disadvantages of centralization
and/or decentralization abound in the literature (Fleming, 1976;
Reynolds, 1977 et al.). Since different authors have different
assumptions and approach the problem somewhat differently, the
arguments are not strictly comparable. For this reason, a table
summarizing most of the pro and con arguments advance in the
literature is condenced in Tables 1-8.
The problems of pure EDP centralization and decentralization
have produced the "common alternatives .. where distributive processing
fits best (Doll, 1977; Atrick, 1976). Some of them are:
1. Operations centralized and system development left to
division - this is an alternative most often adopted in large
organizations producing highly technical products, such as aerospace
manufacturers, with large amounts of scientific and engineering
processing.
37
2. System developme"t centralized and operations
dispersed - · this is an alternative usually found in large business
organizations wit~ geographically dispersed divisions performing
identical functions, none of them of such a nature that very large
computers are required.
3. Central control of equipment acquisitions and central
development of applications common to an entire functional area -
this compromise is generally found in large, geographically dispersed
companies whose divisions and subsidiaries have products representing
a compromise between diversity and commonality.
4. One larger centralized computer plus smaller satellite
computers and remote job entry terminals, and centralized development
augmented by small development groups for unique local needs - a
compromise somewhat simpler than the one above, more appropriate in
smaller and less diversified companies.
5. Centralization of policies for equipment acquisition
and personnel training, some centralized standards, and common
systems for management reporting - this alternative is the most
appropriate to multi-national corporations, where multi-lingual and
multi-cultural factors exist, and different equipment is superior in
different countries.
Cen
tral
ized
Sys
tem
s
*eas
ier
cons
olid
atio
n of
com
pany
w
ide
oper
atin
g re
sult
s *e
ase
of c
ontr
ol
& co
ordi
nati
on
, by
co
rpor
atio
n m
anag
emen
t *e
nhan
ces
corp
orat
e co
nsol
idat
ion
*can
lea
d to
in
tegr
atio
n of
oth
er
adm
inis
trat
ive
func
tion
s *e
asie
r to
im
plem
ent
and
mai
ntai
n s t
anda
r'ds
*sha
red
deve
lopm
ent
cost
s · *
sma 1
1 us
er a
cces
s to
1 ar
ge C
PU
*a
grea
ter
vari
ety
of s
ervi
ces
and
prog
ram
s ca
n be
off
ered
*u
ser
reli
eved
of
mgt
. &
oper
atio
n of
com
pute
r fa
cili
ty
, *ea
sier
to
dir
ect
over
all
use
of
com
putin
g
TABL
E 1
GENE
RAL
AND
ORGA
NIZA
TION
AL C
ONSI
DERA
TION
S
Dec
entr
aliz
ed
Syst
ems
*p
rofi
t &
loss
res
po
nsi
bil
ity
*
fam
ilia
rity
wit
h lo
cal
prob
lem
s *r
apid
res
pons
e to
loc
al
need
s (a
lso
less
for
mal
} *s
peci
al
prog
ram
s an
d se
rvic
es
can
be
tail
ore
d t
o di
visi
on
need
s *e
asie
r con~unication
betw
een
OP a
nd
user
(m
ore
invo
lvem
ent)
· *
11ha
nds-
on11
ex
peri
ence
for
us
ers
pass
ib 1
e . *
mor
e fl
exi
bi 1
i ty
in c
opin
g w
ith
cris
es a
nd. c
hang
es
in p
lan
*b
ette
r se
rvic
e-un
der
user
co
ntro
l *
flex
ibil
ity
in
alig
ning
EDP
w
ith
orga
niz.
ph
iloso
phy
Dis
trib
uted
S
stem
s *f
eeli
ng o
f ex
clus
ive
use
by
user
org
aniz
atio
n *w
hen
usin
g st
anda
rd
equi
pmen
t; th
e de
velo
pm
ent
of
appl
icat
ions
&
tran
sfer
of
pers
onne
l be
twee
n di
visi
ons
easi
er.
*per
mit
s th
e m
ovin
g of
11ex
tra
11
com
pute
r eq
uip
men
t to
div
isio
ns w
ith
an e
xtra
loa
d.
*red
uces
the
num
ber
of
sepa
rate
equ
ipm
ent
\'lh ·i
1 e a
11 o
win
g de
cen
tral
ized
DP
adva
ntag
es
~-----·--------------------------------
--------------------------~~-------------------------~
w
00
TABL
E 2
GENE
RAL
AND
ORGA
NIZA
TION
AL C
ONSI
DERA
TION
S DI
SADV
ANTA
GES
Dec
entr
aliz
ed
Cen
tra 1
i zed
Sys
t_e_m
_s __ ·-----t-----~
5 ter
ns
Dis
trib
uted
Sy
stem
s
*man
agem
ent
prob
lem
s as
soci
ated
w
ith
larg
e st
affs
*p
rone
to
caus
e b
arri
ers
to
acce
ptan
ce
*mor
e li
kel
y t
o ca
use
po
liti
cal
prob
lem
s-
: *hi
gher
ris
k of
fai
lure
1
*mor
e ri
gid
: an
y ch
ange
may
ha
ve s
erio
us r
amif
icat
ions
*r
equi
res
top
mgm
t. in
volv
emen
t *m
ore
vuln
erab
le t
o co
rpor
ate
over
head
re
duct
ion
*man
agem
ent
prob
lem
s as
soci
ated
w
ith
cent
rali
zed
orga
niza
tion
s:
-sta
ndar
diza
tion
-a
ssig
ning
per
form
ance
re
s po
ns ib
i 1 i t
y -a
gree
men
t on
pri
ori
ties
-s
ched
ulin
g pr
oble
ms
-exp
ense
all
ocat
ion
& p
rici
ng
*st
rict
cont
rols
and
sta
ndar
ds
requ
ired
to
prev
ent
dupl
icat
ion
of s
oftw
are
deve
lopm
ent
*no
prof
essi
onal
ED
P m
anag
emen
t *s
epar
ate
equi
pmen
t ac
quis
itio
n st
udie
s &
inte
rcha
ngea
bili
ty
*add
itio
nal
cont
rols
&
stan
dard
s fo
r:
-ens
urin
g co
mm
unic
atio
n be
twee
n si
tes
-saf
egua
rd a
cces
s to
d
istr
ibu
ted
dat
a ba
se
-peo
ple
in D
P re
quir
ed
to h
ave
two
man
ager
s.
*pro
blem
s of
net
wor
k m
anag
emen
t: -i
ncom
e al
loca
tio
n
-exp
ense
all
oca
tio
n
-ass
igni
ng p
erfo
rman
ce
res
pons
i bi
1 i t
y -a
gree
men
t on
p
rio
riti
es
TABL
E 3
COST
FAC
TORS
: AD
VANT
AGE
Cen
tral
ized
Sys
tem
s D
ecen
tral
ized
D
istr
ibut
ed
Syst
ems
Syst
ems
---------------------------------+--------~----------------~------------------~
,------~
*eco
nom
ies
of
scal
e in
mai
n fr
ames
*e
cono
mie
s of
sca
le i
n m
ass
stor
age
devi
ces
*red
uced
rec
ord
stor
age
dupl
icat
ion
*red
uced
sit
e p
repa
rati
on
and
prot
ecti
on c
osts
*f
ewer
ope
rato
rs
requ
ired
*
full
er u
tili
zati
on
of
proc
essi
ng c
apab
ilit
y.
*low
er c
omm
unic
atio
n co
sts
*mod
est
star
t-u
p c
osts
*l
ow
incr
emen
tal
expa
nsio
n co
sts
*hig
her
shar
e of
raw
co
mpu
ting
pow
er
avai
labl
e to
use
r *a
void
s ce
rtai
n u
ser-
com
pute
r co
mm
unic
atio
n co
sts
rela
ted
m
ore
to a
dmin
istr
atio
n th
an
to o
pera
tion
s *
bet
ter
cost
/per
form
ance
*
fast
er r
eact
ion
to n
ew
tech
nolo
gica
l ad
vanc
es
*dep
ends
on
th
e am
ount
of
cen
tral
izat
ion
or
dece
ntra
liza
tion
of
the
syst
em.
Als
o de
pend
s on
th
e ac
tual
ne
twor
k an
d da
ta b
ase
stru
ctu
res
TABL
E 4
COST
FA
CTOR
S: DI
SADV
ANTA
GES
Cen
tral
ized
Sys
tem
s D
ecen
tral
ized
Sys
tem
s D
istr
ibut
ed S
yste
ms '
*may
req
uire
cos
tly
cont
rols
*s
ome
idle
res
ourc
es
*hig
h co
sts
for ex
tens
i~e
conv
ersi
on
*dan
ger
of e
xpen
sive
ove
rhea
d *p
ossi
ble
dupl
i ca
tion
of
*hig
h co
mm
unic
atio
n co
sts
soft
war
e co
sts -
Cen
tral
ized
Sys
tem
s
*gen
eral
sh
orta
ge o
f co
mpe
tent
D
.P.
pers
onne
l *b
ette
r av
aila
bil
ity
in
met
ropo
lita
n ce
nter
s *m
ore
effi
cien
t us
e of
pe
rson
nel
tale
nts
(sp
ecia
li
zati
on}
, *la
rger
and
mor
e ex
pert
poo
l of
con
sult
ants
*b
road
er c
aree
r op
port
unit
ies
-m
ore
attr
acti
ve
posi
tion
. *
high
er s
tand
ards
due
to
mor
e co
mpe
titi
ve s
alar
y le
vels
*p
erso
nnel
tu
rnov
er l
ess
crit
ical
*
:fer
tili
zati
on
*r
otat
ion
of p
erso
nnel
m
ore
natu
ra 1
TABL
E 5
PERS
ONNE
L CO
NSID
ERAT
IONS
: AD
VANT
AGES
Dec
entr
aliz
ed
Syst
ems
*gre
ater
in
tere
st a
nd m
otiv
atio
n at
loc
al
leve
l *
iden
tifi
cati
on
with
th
e m
issi
on
of t
he s
ub-o
rgan
izat
ion
*les
s ri
sk o
f pe
rson
nel
turn
over
*m
ore
oppo
rtun
itie
s to
co
mm
unic
ate
with
(a
nd
tran
sfer
in
to)
lin
e m
anag
emen
t *l
ess
skil
led
per
sonn
el
requ
ired
Dis
trib
uted
Sy
stem
s
*dep
ends
on
th
e am
ount
of
cen
tral
izat
ion
or
dece
ntra
liza
tion
of
the
syst
em.
Als
o de
pend
s on
th
e ac
tual
ne
twor
k an
d da
ta b
ase
stru
ctu
res.
TABL
E 6
TECH
NICA
L CO
NSID
ERAT
IONS
-PRO
GRAM
MIN
G AD
VANT
AGES
Cen
tral
ized
Sys
tem
s D
ecen
tral
ized
D
istr
ibut
ed
.. ~---------
----
----
----
------4----------~Sy~s_t_e_m_s __
____
·---
------r-----S~y_s_t_e_m_s _
____
~'------~
.. *mor
e so
phis
tica
ted
soft
war
e *
bet
ter
serv
ice
to p
rogr
amm
ers
and
user
s -s
yste
m s
oftw
are
can
prov
ide
help
-g
reat
er s
elec
tio
n o
f pr
ogra
mm
ing
lang
uage
s,
debu
g ai
ds,
etc
. *c
an h
andl
e 1 a
rge
prog
ram
s,
no
need
to
bre
ak u
p pr
oble
m
*eas
ier
to
impl
emen
t ch
ange
s in
dat
a ba
se t
echn
olog
y *e
cono
mie
s o
f in
tegr
ated
re
quir
emen
ts
-
*sm
alle
r pr
ogra
ms-
need
to
han
dle
only
on
loca
l si
tuat
ion
*e
asy
to s
atis
fy "
hand
-on"
re
quir
em
ent fo~ te
stin
g p
urpo
ses
*ea
sier
to
add
ne
w ap
plic
atio
ns
and
serv
ices
*f
orce
s m
odul
ar
prog
ram
min
g;
e~sier
to d
ebug
an
d m
aint
ain
*pro
gres
sive
app
roac
h to
in
stal
lin
g
syst
ems
(pro
ject
s br
eak
up
natu
rall
y)
*les
s sp
ecia
lize
d su
ppor
t
*dep
ends
on
th
e am
ount
of
cen
tral
izat
ion
or
dece
ntra
liza
tion
of
the
syst
em.
Als
o de
pend
s on
th
e ac
tual
ne
twor
k an
d da
ta b
ase
stru
ctu
re.
-
Cen
tral
ized
Sys
tem
s
*mul
tipro
gram
min
g li
mit
s pr
ogra
mm
ers
*vir
tual
sto
rage
co
nfl
icts
with
m
odul
ar p
rogr
amm
ing
*mut
ual
inte
rdep
ende
nce
betw
een
jobs
com
plic
ate
both
de
velo
pm
ent
and
oper
atio
ns
TABL
E 7
TECH
NICA
L CO
NSID
ERAT
IONS
: PR
OGRA
MMIN
G DI
SADV
ANTA
GES
Dec
entr
aliz
ed
Sys t
erns
*for
ces
mod
ular
pro
gram
min
g w
hich
is
dif
ficu
lt t
o im
plem
ent
*hav
e th
e pr
oble
ms
with
cu
rren
t m
inis
an
d m
icro
s:
-lit
tle a
ddre
ssab
le s
pace
-n
on-c
ompa
tibi
lity
(ev
en w
ithi
n br
ands
) -n
o ch
oice
s be
twee
n so
ftw
are
vend
ors
Dis
trib
uted
Sy
stem
s
*dep
ends
on
th
e am
ount
o
f ce
ntr
aliz
atio
n o
r de
cent
rali
zati
on o
f th
e sy
stem
. A
lso
depe
nds
on t
he a
ctua
l ne
twor
k an
d da
ta c
ase
stru
ctu
res
Cen
tral
ized
Sys
tem
s
ADVA
NTAG
ES
*red
uced
mea
n va
rian
ce o
n tu
rn
arou
nd
tim
e, w
hich
m
eans
b
ette
r se
rvic
e.
*a g
reat
er v
arie
ty o
f se
rvic
es
and
prog
ram
s ca
n be
off
ered
*l
ess
disr
upti
on w
hen
user
mov
es
(if
both
sit
es
use
sam
e fa
ci 1
i tie
s)
DISA
DVAN
TAGE
S
*sys
tem
sof
twar
e is
com
plex
an
d re
sour
ce c
onsu
min
g
TABL
E 8
TECH
NICA
L CO
NSID
ERAT
IONS
: OP
ERAT
IONS
Dec
entr
aliz
ed
Syst
ems
ADVA
NTAG
ES
*mor
e fa
ult
to
lera
nt
desi
gn
*ea
sier
to
add
new
serv
ice
*les
s sp
ecia
lize
d su
ppor
t ne
cess
ary
*new
er h
ardw
are
tech
nolo
gy o
n av
erag
e
DISA
DVAN
TAGE
S
*use
r m
ay w
ant
to s
tep-
up
to m
ore
eleg
ant
syst
em
*mor
e fr
eque
nt b
reak
dow
ns
Dis
trib
uted
Sy
stem
s
ADVA
NTAG
ES
*hig
her
reli
abil
ity
*
bet
ter
data
com
mun
icat
ions
pe
rfor
man
ce
(few
er
info
rmat
ion
erro
rs)
*fl
exib
ilit
y a
s to
loc
atio
n of
sit
e
*cha
nce
for
bet
ter
secu
rity
DISA
DVAN
TAGE
S
*sys
tem
sof
twar
e is
com
plex
; an
d re
sour
ce c
onsu
min
g
L----------------------------------------------------------~--------------·-------
-------------
46
12. CONCLUSION
To some, distributed systems represent solutions to complex
problems; to others, they create problems too complex for solution.
However, distributed systems (horizontal and vertical) are coming
and both computer professionals and functional management must be
prepared.
Navy Captain Grace Hopper is often quoted (Withinton, 1973)
about a logging operation that uses oxen. When a log is too big
for one ox, they don•t send for an elephant, they use two oxen. So
too with computers according to some distributed systems advocates;
use small or medium scale computers and add another computer as the
work load increases.
If compatible systems (oxen) are used, the interface works.
Suppose, however, to interface an ox and a horse? An incompatibi 1ity -
would exist. A special interface could be design~d and built, but
at significant expense. This illustrates the hardware interface
problem.
Now suppose two oxen were interfaced correctly, but they are
not trained (programmed) to work as a team. A fight could develop
or they could refuse to pull at all. This illustrates software
incompatibility problems.
If both problems are addressed thoroughly, however, a successful
47
distributed system can be attained. In a successful distributed
system of the 1980's one can expect to find an optimal mixture of
horizontal and vertical distributed computers in an on-line total
information system.
It is important to mention the fact that several mainframe
and minicomputer manufacturers like IBM, Digital Equipment, Texas
Instruments and Hewlett-Packard in particular, have started
advocating and supporting the concepts of distributed processing
(Wang, 1976; HP, 1976) outlined in this report. This enhances
the feasibility of successfully implementing such a configuration,
and makes the understanding of the choice process even more critical.
This choice of a system network is dependent on the characteristics
of the applications and how close a network can simulate the
organizational structure, speed and information flow desired.
Finally, the total cost of a distributed system is still higher than
that of a centralized system because of software development and
communications costs. But as hardware costs decreased this last
decade, so software costs are expected to decrease in the next
decade (Eker, 1976; Bremmer. 1976).
48
REFERENCES -.-
Acree, John, and Lynch, Arthur. Dis t:ri buted Processing." 1976, pp. 51-56.
"Ring Network Architecture Supports Data Communications March-April
Anderson, G. A., and Jensen, E. D. "Computer Interconnection Structures ... Computing Surveys 7(December 1975): 197-213.
Bell, Gordon C., and Newell, Allen. Computer Structures: Readings and Examples. New York: McGraw Hill, Inc., 1971.
Benson, James R. 11 The Intelligent Wavehouse. 11 Datamation 22 (September 1976): 107-130.
Black, R. David. 11 Distributed Processing." Datamation 22 (October 1976) : 79-9l.
Bremer, John W. 11 Hardware Technology in the Year 2001. 11 Computer 9 (December 1976): 31-45.
Burns, Christopher J. 11 The Evolution of Office Information Systems ... Datamation 23 (April 1977): 60-65.
Chamberlin, D. D. "Relational Data-Base Management Systems ... Computing Surveys 8 (March 1976): 43-66.
Champine, G. A. "Six Approaches to Distributed Data Bases ... Datamation 23 (May 1977): 69-73.
Cooper, Richard. "Hardware Requirements for Distributed Communications ... Data Communications, April 1977, pp. 51-62.
Curtice, Robert M. 11 Integrity in Data Base Systems." Datamation 23 (May 1977): 64-68.
Digital Equipment Corporation. Introduction to MiniComputer Networks. Boston, Mass., 1974.
Doll, Dixon R. "Relating Networks to Three Kinds of Distributed Functions ... Data Communications, March 1977, pp.37-45.
Ecker, Presger. "Hardware-Software, Then and Now." Computer 9 (December 1976): 54-58.
49
Ens 1 ow, P. H. , Jr. 11 Mul ti processor Organi za ti on - A Survey. 11
Computing Surveys 9 (March 1977): 103-129.
Farber, David J. ..Software Considerations in Distributed Architectu-res ... Computer 7 (March 1974): 31--40.
Flemming, J ... Centralization or Decentralization in Banking ... Datamation 22 (July 1976): 60-67.
Foster, Caxton C. Content Addressable Parallel Processors. New York: Van Nostrand Reinhold Company, 1976.
Foster, John D. 11 Dis tri buti ve Processing for Banking. 11 Da tama ti on 22 (July 1976): 54-58.
Fry, J.P., and Sibley, E. H. 11 Evolution of Data-Base Management Sys terns. 11 Computing Surveys 8 (March 1976) : 7-42.
Hewlett Packard Corporation. The HP Connection. Santa Barbara, California, 1976.
Higbie, L. C. 11 Associative Processors: Computer Design 15 (July 1976):
A Panacea or a Specific? 11
75-87.
Hovey, Richard B. 11 Networks. 11 Data Communications~ May-July 1976, pp. 25-40.
Ka 11 is, Stephen A. Jr. 11 Networks and Distributed Processing ... Mini-Micro Systems 10 (March 1977): 32-40.
Karp, Harry R. 11 Company Networks. 11 Data Conmunications, January-Feb ru a ry 19 7 6 , p p . 19-2 5 .
Keller, R. M. 11 Look-Ahead Processors ... Computing Surveys 7 (December 1975): 177-194.
Kimbleton, S. R., and Schneider, G. M. 11 Computer Communications Networks: Approaches, Objectives and Performance Considerations ... Computing Surveys 7 (September 1975): 129- 17 5.
Korns, Michael F. 11 Halfway to a Relational Data Base . .. Datamation 22 (May 1976): 107-126.
Kuck, D. J. 11 A Survey of Parallel Machine Organization and Programming. 11 Computing Surveys 9 (March 1977): 29-60.
Lea, R. M. 11 lnformation Processing ... Computer 8 (November 1975): 25-33.
50
Lynch, Arthur. "Distributed Processing Solves· Mainframe Problems." Data Communications, November-December 1976, pp. 17-23.
March, Rolf. "Data Processing's Next Stage." Datamation 22 (October_ 1976): 67, 76.
Moore, William G., Jr. 10 (March 1977):
"Going Distributed." Mini-Micro Systems 41-48.
Nielsen, Raynar N. "Distributed-Function Computer Architectures ... Computer 7 (March 1974): 15-19.
Nutt, Gary. "Tutori a 1: Computer Systems... Computer 8 (November 1975): 51-60.
Patrick, Robert L. "Decentralizing Hardware and Dispersing Responsibility ... Datamation 22 (May 1976): 79-90
Reddi, S. S., and Feustel , E. A. "A Conceptua 1 Frame\-lork for Computer Architecture." Computing Surveys 8 (JLine 1976) : 277-300.
Reynolds, Carl H. "Issues in Centralization... Datamation 23 (March 1977): 91-105.
Rodriguez, Juan. "Empirical Behavior in Data Base Systems." Computer 9 (November 1976): 9-17.
Rosenthal, D. B. "The Distributed Data Base Concept... Guide 35 (December 1972): 276-288.
Stone, Harold S., ed., Introduction to Computer Archi tecture. Chicago: Science Research Associates, Inc . , 1975, pp. 318-373.
Taylor, R. W. , and Frank, R. L. "CODAS YL Data-Base Management Sys terns. " Computing Surveys 8 (March 1976 ) : 67-104.
Thurber, K. J., and Wald, L. D. "Associ ative and Par al le l Processors. " Computing Surveys 7 (December 1975): 21 5- 255.
Tsichritizis, D. c .. , and Lochovsky , F. H. "Hi erarchical Data-Base Management... Computing Surveys 8 (March 1976): 105-124.
Wang Laboratories, Inc. Wang on Di s tributed Data Process i ng. Los Angeles, California, 1976.
Withington, Frederic G. 11 Crys t al Ball ing: Trends in EDP Management." Infosystems 20 (January 1973): 20- 26.
51
Yasaki, Edward K. "The Many Faces of the DBA." Datamation 23 (May 1977): 75-81.
Yau, S. S., and Fung, H. S. "Associative Processor Architecture -A Survey ... . -computing Surveys 9 (March 1977): 3-27.