Page images

sense. Paper must be protected against fire and flooding, magnetic tapes against exposure to electromagnetic fields and related hazards. No special precaution is necessary for microfilm, provided

the reels are replaced periodically in updating cycles. Long-term storage of microfilm, however, will require proper temperature and humidity control in the storage area.” (Butler, 1962, p. 64.)

3. Problems of System Networking

3.1 As noted in a previous report in this series: such as those at M.I.T. and the Systems Develop

“Information processing systems are but one ment Corporation are linked by digital transmission facet of an evolving field of intellectual activity channels. The difficulties involved in establishing called communication sciences. This is a generic computer networks appear not to be difficulties of term which is applied to those areas of study hardware design or even of hardware availability. in which the interest centers on the properties Rather, they appear to be difficulties of social and of a system or the properties of arrays of symbols software organization, of conventions, formats, which come from their organization or structure and standards, of programming, interaction, and rather than from their physical properties; that is, communication languages. It is a situation in which the study of what one M.I.T. colleague calls the there now exist all the component hardware facilities problems of organized complexity'.” (Wiesner, that can be said to be required, yet in which there 1958, p. 268).

do not now exist any general-purpose networks The terminology apparently originated with capable of supporting stage-three interaction.” Warren Weaver. Weaver (1948) noted first that the (Licklider, 1967, pp. 5-6). areas typically tackled in scientific research and “The state of affairs at the end of 1966 can be development efforts up to the twentieth century summarized as follows. Multiaccess-system applicawere largely concerned with two-variable prob- tion techniques and user-oriented subsystems have lems of simplicity; then from about 1900 on, power- been developed to only a relatively primitive level. ful techniques such as those of probability theory Far too many people, particularly those with a and statistical mechanics were developed to scientific-application bent, still hold the shortdeal with problems of disorganized complexity sighted view that the real value of large, multiaccess (that is, those in which the number of variables systems lies in their ability to simultaneously is very large, the individual behavior of each present a powerful 'private computer' to each of of the many variables is erratic or unknown, but

several tens or hundreds of programmers. Nearly the system as

a whole has analyzable average all of the systems actually in operation are used in properties). Finally, he points to an intermediate this way. Application-oriented systems that free the region "which science has as yet little explored user of most, or all, of his concern with details of or conquered" (1948, p. 539), where by contrast computer-system characteristics and conventional to those disorganized or random situations with programming are coming much more slowly. Their which the statistical techniques can cope, the development has been inhibited by, among other problems of organized complexity require dealing things, the temporary plateau that has been reached simultaneously with a considerable number of in basic multiaccess system technology.” (Mills, variables that are interrelated in accordance with 1967, p. 247). organizational factors.

3.3 "The analytical tools are simply not avail. 3.2 "Organizational generality is an attribute

able ...” (Baran, 1964, p. 27). of underrated importance. The correct functioning

“The essence of rational benefit-cost analysis is of on-line systems imposes requirements that have

the tracing of indirect as well as direct effects of been met ad hoc by current designs. Future system

programs and the evaluation and summing of these designs must acknowledge the basic nature of the

effects. Typically, the methodology for tracing all problems and provide general approaches to their

but the most obvious linkages is entirely lacking or resolution." (Dennis and Glaser, 1965, p. 5).

fails to use the relevant information." (Glaser et al., "Diversity of needs and divisibility of computer 1967, p. 15). resources demand a much more sophisticated multi- “The problem associated with providing the interplexing strategy than the simple communication

connection of a network of processors is a major case where all users are treated alike.” (David, 1966,

one.” (Estrin and Kleinrock, 1967, p. 92).

"Solving the data base management problems of a “As we turn toward stage three, the stage char- distributed net work has been beyond the state of acterized by the netting of geographically distrib- the art." (Dennis, 1968, p. 373). uted computers, we find ourselves with a significant “Although techniques for multiplex communica. base of experience with special-purpose computer

tions are well developed, we are only beginning to networks, but with essentially no experience with learn how to multiplex computers.” (David, 1966, general-purpose computer networks of the kind that p. 40). will come into being when multiple-access systems

“The formalism of hardward/software system

p. 40).


controlling the use of such systems. These problems have evidenced themselves in computer scheduling, program capability constraints, and the allotment of auxiliary storage.” (Linde and Chaney, 1966, p. 149).

. "A network has to consider with great care the many possibilities of user access which approach more and more the vast possibilities and intricacies of direct human communication.” (Cain and Pizer, 1967, p. 262).

“Increased attention needs to be placed on the problem of techniques for scheduling the many

with their different priorities." (Bauer, 1965, p. 23).

3.6. "Much of the design effort in a messageswitching type communications system goes into the network which links the terminals and nodal points together. The distribution of terminals can be shown, the current message density is known, and programs exist to help lay out the network. With most interactive systems this is not the case.” (Stephenson, 1968, p. 56).

3.7 “In looking toward computer-based, computer-linked library systems, that have been proposed as a national technical information network, studies of perceived needs among users are likely to be of very little use. Instead it would

to be more appropriate to initiate smallscale experiments designed to produce, on limited basis, the effects of a larger-scale system in order to determine whether such experiments produce the expected benefits." (Schon, 1965,


management is just beginning to take shape. SAGE is a landmark because it worked in spite of its immense size and complexity.” (Aron, 1967, p. 50).

“The system engineer presently lacks sufficient tools to efficiently design, modify or evaluate complex information systems." (Blunt, 1965, p. 69).

3.4 “The design and analysis problems associated with large communications networks are frequently not solvable by analytic means and it is therefore necessary to turn to simulation techniques. Even with networks which are not particularly large the computational difficulties encountered when other than very restrictive and simple models are to be considered preclude analysis. It has become clear that the study of network characteristics and traffic handling procedures must progress beyond the half-dozen switching center problem to consider networks of dozens of nodes with hundreds or even thousands of trunks so that those features unique to these large networks can be determined and used in the design of communications systems. Here it is evident that simulation is the major study tool.” (Weber and Gimpelson, 1964, p. 233).

“The time and costs involved make it almost mandatory to prove the 'workability and feasibility of the potential solutions via pilot systems or by implementation in organizations or associations which have some of the characteristics of the national system and which would therefore serve as a model or microcosm of the National Macrocosm." (Ebersole, 1966, p. 34).

“A report by Churchill et al., specifically recognizes the need for theoretical research in order to build an adequate foundation on which to base systems analysis procedures. They point out that recent computer developments and particularly large computer systems have increased the need for research and the body of data that research can provide in such areas as data coding and file organization." (Borko, 1967, p. 37).

3.5 “The coming importance of networks of computers creates another source of applications for . . . multiple-queue disciplines. Computer network disciplines will also have to be dependent on transmission delays of service requests and jobs or parts of jobs from one computer to another as well as on the possible incompatibilities of various types between different computers. The synthesis and analysis of multiprocessor and multiple processor network priority disciplines remains fertile area of research whose development awaits broad multiprocessor application and an enlightening experience with the characteristics of these disciplines." (Coffman and Kleinrock, 1968, p. 20).

“We still are plagued by our inability to program for simultaneous action, even for the scheduling of large units in a computing system.” (Gorn, 1966, p. 232).

As computer time-sharing systems have evolved from a research activity to an operational activity, and have increased in size and complexity, it has become clear that significant problems occur in


p. 34).

3.8 “Transmitting data collection systems can assume a wide variety of equipment configurations, ranging from a single input unit with cable-connected recorded to a farflung network with multiple input units transmitting data to multiple recorders or computers by means of both common-carrier facilities and direct cable connections. Probably the most important parameter in planning the equipment configuration of a system is the maximum number of input stations that can be connected to a single central recording unit." (Hillegass and Melick, 1967, pp. 50-51).

3.9 Licklider stresses the importance of “coherence through networking” and emphasizes: “On the average, each of n cooperative users can draw n-1 programs from the files for each one he puts into the public files. That fact becomes so obviously significant as n increases that I can conclude by concluding that the most important factors in software economics are n, the number of netted users, and c, the coefficient of contributive cooperativeness that measures the value to his colleagues of each user's creative effort.” (Licklider, 1967,


p. 13).

[ocr errors]

“The circumstances which appear to call for the establishment of physical networks (as opposed to logical networks) are generally: “1. The existence of special data banks or

special collections of information located at a single institution but useful to an audience

geographically dispersed. “2. The inadequacy of general data banks or

general collections of information to meet local needs where remote resources can be used in a complementary fashion to fulfill the

need. “3. The centralization of programming services,

processing capabilities or scientific resources

with a geographically dispersed need. “4. The need for interpersonal (including inter

group) direct communication. This includes

teleconferencing and educational activities. “5. A justification on economic, security or social

grounds for distribution of responsibility for load sharing among organizations or geo

graphical regions." (Davis, 1968, pp. 1-2). “In certain areas, such as law enforcement, medicine, social security, and education, there is a need for joint Federal-State computer communications networks which can apply new technology to improving the management of major national programs in these areas.” (Johnson, 1967, p. 5).

“It has been suggested that a principal advantage to be gained from computer networks is the ability to distribute work evenly over the available installations or to perform certain computations at installations particularly suited to the nature of the job.” (Dennis, 1968, p. 374).

“The time-sharing computer system can unite a group of investigators in a cooperative search for the solution to a common problem, or it can serve as a community pool of knowledge and skill on which anyone can draw according to his needs.” (Fano and Corbató, 1966, p. 129).

“Within a computer network, a user of any cooperating installation would have access to programs running at other cooperating installations, even though the programs were written in different languages for different computers. This forms the principal motivation for considering the implementation of a network." (Marill and Roberts, 1966, p. 426).

3.10 “The establishment of a network may lead to a certain amount of specialization among the cooperating installations. If a given installation, X, by reason of special software or hardware, is particularly adept at matrix inversion, for example, one may expect that users at other installations in the network will exploit this capability by inverting their matrices at X in preference to doing so on their own computers." (Marill and Roberts, 1966,

[ocr errors]

c. all operations may be publicized.” (Brown

et al., 1967, p. 209). 3.12 “The operator's charges to clients must be in fair proportion to the usage made of installation resources (processing, time, storage occupancy, etc.). Therefore adequate records must be kept of resource use.” (Dennis, 1968, p. 375).

"A principal (individual or group of individuals) is charged for resources consumed by computations running on his behalf. A principal is also charged for retention in the system of a set of computing entities called retained objects, which may be program and data segments ...” (Dennis and Van Horn, 1965, p. 8).

“The equitable allocation of space and time by administrative fiat. This is probably an overwhelming problem in a network since the predicting of communication paths and computing facilities required by any user would be quite unwieldy. ...

“Thus a more elaborate scheme seems to be necessary-one whose rates are proportional to the value of the service. This would be modified in a network because the rate for the same kind of service may vary among the installations." (Brown et al., 1967, p. 212).

“Built-in accounting and analysis of system logs are used to provide a history of system performance as well as establish a basis for charging users. (Estrin et al., 1967, p. 645).

3.13 "Questions of technical feasibility and economic value are not the sole determinants of the computer utility. The development of the computer utility may be influenced by norms, or lack of norms, about the confidentiality of data. At the moment there do not seem to be any clear standards of good practice; perhaps there was less need before technology greatly increased

greatly increased capabilities for handling data." (Jones, 1967, p. 555).

“It should be easy and convenient for a user to allow controlled access to any of his segments, with different access privileges for different users.” (Graham, 1968, p. 367).

3.14 "The designer must decide whether he will provide the high-speed service to all users, to provide the service that the majority request, and leave the minority to fend for themselves, or to provide the degree of speed needed in each case, but no more. Ideally, he should know the entire distribution of response time requirements. It is even desirable to know how the arrival of these queries will be distributed in time throughout the day. In attempting to meet requirements, he must consider what is actually being retrieved in any stated response interval. Does the user want hard copy or will he be satisfied with citations or index records? The engineering problems associated with high-speed retrieval of hard copy from files can be formidable. If a conversational, or browsing, mode of search is used in which the searcher uses a succession of queries, do we aim to minimize his total search time, or only to give him immediate

p. 426).


"An interconnected network would make it possible for the top specialists in any field to instruct anyone within the reach of a TV receiver. (Brown et al., 1967, p. 74).

3.11. "For initial network trials the advantages of an open system are:

a. ease of programming, b. services for all users,

p. 191).




response to each single query?” (Meadow, 1967, oriented problems, four major system equipment

performance factors must be evaluated: "Precedence is computed as a composite function 1. Real time processing capability of central of:

processor(s) 1) the ability of the network to accept additional

2. Core memory size provided in central proctraffic:

essor(s) 2) the 'importance of each user and the 'utility'

3. Bulk storage size provided

4. Limitations on real time of his traffic;

to bulk 3) the data rate of each input transmission

storage.” (Birmingham, 1964, p. 38.) medium or the transducer used;

3.16 "It seems imperative that EDUCOM ... 4) the tolerable delay time for delivery of the establish certain technical standards and operatraffic." (Baran, 1964, p. v).

tion procedures which each state or regional group “Many separate low-data-rate devices time-shared

must meet before they can be interconnected. • or concentrated into a single high-data-rate link

These standards should apply to digital transpermit better averaging, as compared to a few

mission, telephonic communications, and tele

vision correspondingly-higher-data-rate users. But,

.” (Brown et al., 1967, p. 54). many of the high-data-rate users ‘get in' and 'get “One final thought about integration. Integration out' fast, they have a short holding time. This helps is facilitated by the standardization of equipment, the averaging process. To be precise in this compu

processes, and languages. However, standardizatation, a better understanding of the number of tion in command-and-control systems must be users, their use statistics, and the network charac- considered in the light of the evolutionary nature teristics appears mandatory.

of these systems. First, standardization should "... The mixed requirement that, while we

be based upon those elements which are missionwish to give priority treatment to the higher

independent; that is, the elements standardized precedence traffic of equal network loading, we

should be general-purpose in nature. Secondly, must also satisfy the goal that we preserve a mini

the standardization of system elements should mum transmission capability for the lower-prece

be modular; that is, it should be possible to add dence traffic.. Thus, instead of a blanket rule that

other elements to them in order to modify or all traffic of a given precedence grade will be trans

increase the capabilities of the system. If system mitted before handling the next lower precedence

elements are standardized at too low a level of grade, we choose to use the time ratios of these

aggregation, the system's speed of response is precedence categories to act as a preference

increased, but its flexibility is reduced. If, on the weighting factor.” (Baran, 1964, pp. 30, 33).

other hand, elements are standardized at the

higher levels of aggregation, provided these are “An added complication may be introduced not higher than the level of the designer's probin the form of a hierarchy of precedence classifica- lem, flexibility is increased, but there is an accomtions. This can be a very useful feature of the com- panying reduction in the system's speed of remunications system, allowing important messages to sponse. It is this trade-off between flexibility avoid delay by by-passing a string of messages of and speed of response that makes the standardizarelatively low urgency. But it adds an extra dimen

tion problem such a difficult one.” (Jacobs, 1964, sion to the message queue, requiring separate list- pp. 41-42). ings for each precedence. This system can go beyond governing of the order of transmission, and can

"In general the standards of distributed-control allow high-priority messages to interrupt others

systems are standards built around each class of during their transmission. In such a system,

job for each level of job for each unique function of

, message switching has an advantage over circuit

the system. Procedures and languages need not switching, in that an interrupted message can be

be standardized across job levels or across functions. automatically retransmitted as soon as possible,

Minimum standardization does not, however, imply

the complete freedom of each functional unit to with no further action by the sender. But the pos


select idiosyncratic communication sibility of interruption necessitates that the entire contents of a message be retained in storage

bizarre formats. Such matters as codes, formats, until its last transmission is completed.” (Shaf

file structures, vocabularies, and message syntax ritz, 1964, p. N2.3–3).

are all aspects of performance programs, and the

library of these program building blocks, from which "The difference between control strength and

any information-processing job can be built, is priority is that control strength is used for defining

bounded from above. Executive control over the interrupt classes (an interrupt class is the set

limits of the library establishes the boundaries of of all requests with the same control strength),

the range of alternatives available at any organizawhile priority is used for ordering requests within tional level. This is standardization of a sort, but it the same interrupt class.” (Dahm et al., 1967,

allows considerably more flexibility than the standp. 774).

ardization generated by a rigid set of specifications 3.15 "In real time data communications to be applied across functions and up and down the


hierarchy of information-processing jobs." (Bennett, 1964, p. 107).

3.17 “A built-in system for user feedback would be essential in determining near-future needs and current inadequacies of the network.” (Brown et al., 1967, p. 216).

“It was considered that it might be useful to have all users of materials feed back their evaluations, which could be analyzed statistically for consideration of the next user." (Brown et al., 1967,

[ocr errors]

p. 63).

[ocr errors]


“The least expensive method of organizing a science information system network appears to be on a regional basis with the centralization of acquisition input processes being undertaken in a central clearing house." (Sayer, 1965, p. 142.)

3.20 “In the network concept, then, the technical information centers would be linked by the traffic ro ing centers. Each would become dependent upon the other with both responding to the law of supply and demand, service and customer satisfaction, and continued viability based upon justification of existence through performance. (Vlannes, 1965, p. 5).

“Other choices in the spectrum may include that of a network of information centers in which each community performs and contributes to the advancement of knowledge in accordance with its capabilities. Of course, a network must impose a series of constraints in order to operate, but it also allows for the flexibility that a rigidly structured system cannot accommodate. A network also fosters a sense of competition in which each community must ever strive to re-orient itself in order to survive and progress in its changing environment. In addition, each must become sensitive to the changes in the other communities in order that it may react, re-evaluate and adapt to the net set of goals that are inevitable." (Vlannes, 1965, p. 4.)

“In order to gain control over the accountability data, a telephone switchboard was added to the system. With the formalization of the terminal network, the concept of operation changed from a central computer with satellite terminals to the concept of a central terminal network with satellite computers.” (O'Sullivan, 1967, p. 169).

3.21 “Many of the larger systems must also take into account the requirements for providing machine-readable output for use in a decentralized network of search centers. The designer must remember that other users will place constraints on the parent system. It must be remembered that a change to the central system has multiple effects on the various members of the decentralized network. Good system documentation will be essential in providing programs to the local search centers. A constant training requirement will also be imposed upon the central system, and technical liaison must be maintained with all users in the network. Effective file maintenance procedures must be developed well in advance of implementation of the decentralized system. Changes and updatings to the central nle will occur frequently, and an adequate mechanism must be available for insuring that these same changes are made to all files in the field.” (Austin, 1966, p. 245).

“Locating the point of minimum sufficient centralization for a system may call for a some what atypical philosophy of system design, a philosophy not commonly held by theoreticians on the subject but often implicit in the daily design practices of the engineers and logicians involved in the actual specification of system details. That

“Provisionally we characterize a network by: A. Remote and rapid services regarding selec

tion, acquisition, organization, storage, retrieval, and processing of information and

procedures in current files B. Feedback to the

1. Originator or organizer of the informa

tion (hence there would be a Community of users improving a common store of

materials and procedures). 2. Supervisor of the network services

(hence the system would be adaptive to the needs of the users).” (Brown et al.,

1967, pp. 49–50). "In designing a priority handling system, we should never permit ourselves to believe that we have more (or less) usable communications capability than we really have. This implies network status control feedback loops." (Baran, 1964, pp. 17-18).

3.18 “To facilitate system scaling, reliability, and modularity, many multi-processor operating systems are designed to treat the processors as homogeneous system resources. Hence, there is no 'supervisor' processor, each schedules and controls itself. To prevent critical races and inconsistent results, only one processor at a time is permitted to alter or examine certain shared system data bases; all other processors attempting simultaneous

are locked-out. This phenomenon is not strictly limited to homogeneous processor systems, similar requirements apply to any multi-processor scheme utilizing shared data bases.” (Madnick, 1968, p. 19).

3.19 "Some general observations may be of interest. There are indications that the cost of operating an information system network, organized along subject lines, varies little with change of process allocation within the system. Whether all acquisition and input processes are carried on in a center clearing house or distributed in some logical manner among the service centers does not appear to make a significant difference in cost. On the other hand, centralization in the regionally organized system becomes imperative if excess operating costs are to be avoided. In a system organized to serve users on a project basis, there is an indication of some economy of operation being achieved by complete decentralization.” (Sayer, 1965, pp. 141-142.)



« PreviousContinue »