Referring to Figure 1, we see that in some cases the processing operations of Box 4 result in direct outputs (Box 5) and/or archival storage (Box 6) as in the case of preservation of input records for subsequent reconstruction if required. In most cases, however, processing service requests are received from clients of the system, from the system itself, or by implications in the incoming data received in Box 2. In parallel with the input data transformations shown in Box 3, we may find reductions or translations of processing service requests to forms or formats that can be processed by the system in Box 8, which may also involve preprocessing operations that are required by priority and scheduling considerations in multiple access systems. Processing specifications developed in Box 9 may be exemplified by controls imposed by measures taken to provide both client and system protection. These specifications, and others, will then be matched (Box 10) to the performance capabilities of the processing system itself and used to control the actual processing operations.

2.1. General Considerations

Processor system planning and managementi activities must effectively interlace the requirements of multiple inputs of service requests (Box 7),.. the handling of queuing and job scheduling priorities, (Box 8), the provision of suitable means for both client and system protection (Box 9), the require-; ments of matching the processing service requests; to the main processing system capabilities by appropriate supervisory and executive control routines (Boxes 10 and 4), and the orderly flow of both programs and data to and from various levels of storage (Box 6).

In Figure 2 we show those functions of Figure 1. that are most directly involved in the requirements for efficient planning and management of information processing systems as such, together with indications of areas of current and continuing R & D concern. We may consider more particularly here the system design requirements for computerbased multiple access or "time-sharing" systems.

"Time-sharing" is the most prevalent and popular terminology for the concept of information processing systems used by many different clients, with effectively simultaneous response to their diversified processing service requests, with at least the apparent effect that the processing facilities are "on-line" to them when they desire to use them and with certain sharing of both data and programs among different users.2.1

In fact, some of this terminology is misleading, and the processing system capabilities are not so much shared as divided among many users, with considerable swapping of system resources between them, and with only intermittent access between any individual user and the processor-storage system and its procedures. Thus, we generally prefer to use the more comprehensive term "multiple-access system" to mean "a system in which multiple independent users are provided service by the computer via remote consoles which asynchronously communicate with the computer on a demand basis." (Morenoff and McLean, 1967, p. 19). In general, the systems so-called do not provide time-sharing capabilities, but they do offer system-sharing capabilities.

2.2. Management of Multiple-Access Systems

Considerable current progress can certainly be shown in the development and experimental use of computer-based multiple-access or "timeshared" systems.2.1a Nevertheless, many difficult problems remain, including questions of responsive scheduling, effective monitoring, fail-safe private. file or data protection, languages of access, and efficient supervisory control.2.2 A complex intermixture of hardware, software, and human behavioral factors is typically involved.

It is particularly to be noted that major R & D efforts may be required to implement the diversified functions of man-machine reactive procedures with appropriate degrees of client convenience, system efficiency, and economy.2.3 Licklider suggests further that "industry has not devoted as much effort to development of devices and techniques for on-line man-computer interaction as it has to development of other classes of computer hardware and software." (Licklider, 1965, p. 66.) Those responsible for R & D program planning in the computer and information sciences should therefore be urgently concerned with the problems raised by the potentials for increasing use of multiple-access systems, generally.

2.2.1. General Considerations and Examples Phenomenal growth is promised in the use of such multiple-access systems over the next few years.2.4 Continuing difficulties are predicted, however, with respect to such areas as adequate programming languages, the appropriate design of small inexpensive but versatile consoles for personal use, and the provision of effective privacy

and protection facilities. From the system design and management point of view, the challenges are to take an overall planning approach, to look toward the integration of many processes that are now distinct, to seek maximum gains from all multipurpose processing opportunities that are available, and to undertake a drastic rethinking of possible trade-offs between benefits and costs.2.4a

Sutherland (1965) considers six varieties of on-line information processing systems: (1) processing control systems, where in manufacturing applications the potentialities range from feedback control to the optimization of profit expectancies; (2) inquiry systems, such as those currently used for airline reservation purposes; (3) specialized on-line systems for specific military applications but also for engineering and design; (4) on-line programming systems; (5) on-line problem-solving systems; and (6) on-line instrumentation designed to bring a better understanding of the interplay of the programs and data within the computer.

As examples of experimental developments and applications of multiple-access systems whose facilities are shared by different users, we may cite first the case of mechanized documentation operations and secondly that of on-line problem-solving. The "reactive typewriter plan", proposed by Mooers at least as early as 1959.2.5 foresaw not only interlibrary cooperation and interchange of records, but tie-ins to a remotely located library or information center for the individual user as well. Inputs from remotely located typewriters or consoles via teletype, telephone, and other communication links, with or without secondary mode transmission via voice channels, are the heart of remote-access timesharing systems already in experimental operation.

Among these experimental systems we note in particular those that provide responsive and selective access to files of bibliographic data involving scientific and technical literature. A first example is that of the use of the M.I.T. Compatible TimeSharing System (CTSS) in Project MAC.2.6 This has been used both by Kessler and in the experimental SMART system developed by Salton and associates at Harvard.2.7 Kessler's application involves machine access, in a variety of selection-retrieval modes, to a corpus consisting of bibliographic references to periodical literature in the field of physics. Recent developments include the implementation of a "more-like-this" request feature and a provision for delayed service.2.8 Brown 1966 describes the use of this system for updating a book on the basic data of plasma physics. For the future, Wooster in a 1968 report proposes the Bibliographic On-Line Organized Knowledge System, or BOOK.

At the System Development Corporation, the Q-32 time-sharing system is used both for experimental application of question-answering procedures to reasonably terse, relatively well-structured data (robbery reports of the Los Angeles Police Department) and for more generalized natural language text processing with respect to the Golden Book children's encyclopedia.2.9

The BOLD (Bibliographic Organization for Library Display) Project initiated at SDC by Borko, involves a number of user stations equipped with CRT display console, light pen, and teletypewriter. The light pen is used to designate document categories (all members of which are displayed), to specify data to be furnished in hard copy form, and to reject data displayed during document search. The keyboard is used to enter desired author or index tag selection criteria. (Borko, 1965; Carter et al., 1965, pp. 62-63).2.10

Lockheed Missiles and Space Company's on-line document reference retrieval system was designed for the in-house report collection of some 100,000 items as covered in the MATICO (Machine-Aided Technical Information Center Operations) system. Search and retrieval procedures, on-line, are based upon such access points as personal and corporate author names, keywords in title, subject headings as assigned, report and contract numbers, and date of publication.2.11 Search restrictors such as announcement media may be employed and associational displays of index terms used in the system may also be provided.2.12

Further details are provided by Summit (1967), who describes the development of CONVERSE into DIALOG, a system developed to investigate experimentally the effectiveness of a user-directed language designed to provide flexible means for reference retrieval.2.13 A resource-allocation scheme for this Lockheed multiple-access library reference retrieval system is discussed by Reiter (1967).

Other relatively recent examples include applications at the Moore School of the University of Pennsylvania, such as Project CIDS (Chemical Information Data System), which deals with formulas, names and properties of chemical compounds that match a client's query, and Project Vector which provides library information retrieval services; 2.14 an experimental system which has been developed by Bunker Ramo for NASA; 2.15 the augmented catalog of project Intrex 2.15a and SAFARI, an on-line text processing system at the MITRE Corporation.2.16 It has already been claimed that "there is a general feeling that on-line retrieval is the next major development in information retrieval, and represents the retrieval system of the future." (Drew et al., 1966, p. 341).

It will be noted that many of the proposed applications of such systems imply on-line problemsolving capabilities on the part of the client in the special case of retrospective literature search.2.16a More generally, on-line problem-solving possibilities are claimed to provide one of the most challenging areas for the further improvement and increased use of multiple-access systems.2. 2.16b

Such general systems will need to meet system design requirements especially suited to the creative user. They require consideration of the convenience to such clients in learning to use, in using, and in modifying their use of the capabilities offered.2.17 In particular, such systems may be designed to some extent as 'teaching' machines 2.18 and as

systems that provide for "mutual training" of the man and the machine.2 2.19 Sutherland suggests tha on-line problem-solving systems "will require techniques for pattern recognition, process control and heuristic programming, and will unite then meaningfully." (1965, p. 9.)

In the general case, however, these potential are closely related to far-ranging prospects fo graphical manipulation and display, especially fo applications to machine-aided design operations These prospects, however, will be considered in later section of this report (5.1.2). Let us nex consider here, then, the preprocessing of clien service requests in terms of job scheduling, queuing and priorities.

2.2.2. Scheduling, Queuing, and Priorities in Multiple' Access Systems

In general, it is noted that "computer time-sharin attempts to phase the simultaneous execution o two or more programs so that the execution of any one of the programs occurs during the natura. dormancy period of the others" (Brown, 1965, p. 82, and that, "typically, a particular user's progran, will be allowed to use a processor for a period o time, will be stopped so that another user's progran run, and then at some later time will b continued from the point where it was stopped. (Scherr, 1965, p. 1).2.19a It has been appropriately observed that effective time-sharing thus involve time-slicing.2.20

can

This is a problem area in which both theoretica assumptions and pragmatic considerations baseq upon relatively limited experience to date require further detailed investigation and evaluation experimental exploration of alternatives, and stud of the design requirements for judicious balancing of hardware and software contributions to an effec tive system solution.2.21

The responsive scheduling of both client access and job-processing requirements will necessarily involve effective queuing procedures, routing service and priority allocations, and on-line monitor, ing of system operations (whether or not accounting and billing requirements are als involved.) Coffman and Kleinrock provide a 1968 survey of priority policies and scheduling method in use and, in addition, they discuss some of the means by which various users may try to beat the system. Lampson (1968) notes further problems of management-control enforcement.2.21a

Some number of different user-control mode! should be typically available. For example, in term of commerical service-center operations, Adams, (1965, p. 487) suggests that: "The scheduling methoc! used in the KEYDATA system. . . allows con versational users to reserve a level of service fo minutes or hours at a time and guarantees tha they will receive at least the agreed level of services on a minute-by-minute basis." In other situations clients may wish to control whether their jobs are to be processed by remote batch operations or in a conversational mode.2.22

In the EDUNET planning conferences (sponsored by the Inter-University Communications Council), it was pointed out that both priority allocations (including pre-emptive priorities) and reservation mechanisms would be needed in order to assure different classes of users their desired levels of service.2.23 In particular, Caffrey stressed the : following: "Priority is not a simple matter; there are several kinds and issues:

entry into the system (being able to plug in); preferential treatment in the stack or queue; quality of response;

efficiency of operation;

user-priority vs. job-type priority;

levels of hardware, including storage capacity." (Brown et al., 1967, p. 214).

Early evidence of R & D concern in many of these areas was presented at the 1963 Fall Joint Computer Conference (Aoki et al., and Critchlow, for example). Of the early JOSS (Johnniac Open-Shop System) developments, beginning in 1963-1964, Bryan reports that considerable effort was devoted to instrumentation measures to record not only the use of the system as a whole but also the characteristics of usage by different users.2.23a Kleinrock (1966) discusses the theory of queues in relation to experience of usage in time-shared computer systems. Some SDC developments include those of Coffman and Kirshnamoorthi (1964), Fine et al. (1966), Krishnamoorthi and Wood (1965) and Totschek (1965).

Where there are many system clients for whom effectively simultaneous access should be provided, appropriate scheduling and queuing provisions must be made. To date, such provisions typically range from first-come-first-served and round-robin arrangements (after a given processing interval, if the first client's problem has not been completed, his programs, data, results to that point, and status and re-start information are transferred out of the active part of the system. Then the processingservice requests and data of the next client in line are swapped in, the system returning to the first client only after all waiting clients have had an equal opportunity for processing service) to procedures based upon considerations of length of running time required (e.g., short jobs first), with or without priority-interrupt considerations.2.24

In the latter case, overall system efficiency may have a built-in self-improvement facility through the training or disciplining of the typical client. Scherr (1965, p. 105) suggests that: "it seems that scheduling, if properly executed, could be used to 'mold' the users to some extent by assigning priority on the basis of job type, program size, program running time, user think time, etc., etc. Then the user, in trying to 'beat the system', will tend to conform to the image of what the writers of the scheduling program considered to be the ideal user." Further, "if the scheduling procedure gives low priority to a user's program because of one of

its characteristics (e.g., program size), users seem to try to eliminate these objectionable' features from their programs and interaction usage."

On-line access scheduling may be arranged on either an a priori or a dynamic basis.2.25 Typical system design and operation questions also arise as to scheduling and queuing with respect to multiple system components and as to the number of multiple components necessary to maintain effective service for the clientele.2.26

Areas of specific needs for further R & D investigation include the following types of questions: (1) Whether the specific time interval (or “quantum") allowed to the active client in the waiting line allows, for most clients, an acceptable delay-tolerance and, at the same time, good system utilization? 2.27

(2) Whether mean-response-time is a good criterion of either probable client-acceptance or of system utilization? 2.28

(3) Whether overhead time-and-cost requirements have been effectively assessed? 2.29

(4) Whether comparison of program sequences and data units is sufficiently systematic to decide which priority choices will best contribute to effective utilization? 2 2.30

In general, it has been claimed that "the major issues involved in time management are those of selecting a queue discipline, determining optimum quantum size, and dealing with system overhead." (Schwartz and Weissman, 1967, p. 264.) On-line monitoring, it should be noted, will often be required in order to adjust for different input and service requirements "mixes" and to guard against such phenomena as a number of different clients "getting into phase with each other." 2.31

Then there is the difficult question of scheduling in multiprocessor, multiprogrammed 2.31a networks and, more generally, in systems incorporating the concept of the "information processing utility".2.32 Stanga (1967) discusses the performance advantages developed for the Univac 1108 multiprocessor system in terms of multiple processing units and multiple access paths to a variety of input-output and peripheral storage devices. Manacher (1967) considers the special problems of multiprocessor control in the case of a 'hard' real-time environment - that is, one where there are rigid timing bounds (both start-times and deadlines) upon system performance.

Dennis and Van Horn (1965, pp. 3-4) discuss various major characteristics of multiprogrammed, multiprocessor systems in terms of the following factors: (1) computational processes are run concurrently for more than one user, (2) many processes must share resources in a flexible way, (3) individual processes post widely varying demands for computing resources in the same time period, (4) separate processes make frequent references to common information, and (5) the system itself must be capable of flexible adjustments to changing require

ments.