A Single System Image: An Information Systems Strategy

viProfessional Paper Series, #1

The ProfessionalAssociation

for Computingand Information

Technologyin Higher Education

By Robert C. Heterick, Jr.

C USEA

A Single System Image:An Information Systems Strategy

By Robert C. Heterick, Jr.

Professional Paper Series, #1

The Professional Associationfor Computing and Information Technology

in Higher Education

C USEA

ii

Copies of this paper are available to staff of CAUSE member institutions at $8 per copy, to non-members at $16 per copy. Orders should be pre-paid and sent to:

CAUSE Publications737 Twenty-Ninth Street

Boulder, Colorado 80303(303) 449-4430

Copyright © 1988 by CAUSE, The Professional Association for Computing and InformationTechnology in Higher Education.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system,or transmitted in any form or by any means without prior written permission from CAUSE.

Printed in the United States of America.

This paper was produced on an Apple Macintosh Plus computer and LaserWriter Plus printerusing PageMaker software. Hardware and software were contributed to the CAUSE NationalOffice by Apple Computer, Inc., and Aldus Corporation.

iii

PREFACE All men can see the tactics whereby I conquer, but what none can see is the strategy out ofwhich victory is evolved.

Sun Tsu (3000 B.C.)

This paper addresses both strategy and tactics. It attempts to weave together a previouslypublished article on the subject of strategic planning for information systems with a reasonablydetailed description of the products needed to purvey an institution’s information resources asthough they were delivered from a single, integrated system. Several of the implicitly acceptedpremises for the current state of higher education information systems are set forth. The impactof technological change in the computer and communications fields is reviewed and its impacton these premises is assessed. The relationship of the institution’s own strategic goals to thedevelopment of an information infrastructure is also investigated.

The need for a strategy, as opposed to a set of goals, is discussed in light of such rapidlychanging technological parameters. A guiding strategy that will take the institution through thenext five to ten years is developed by raising a series of typical questions that will need to beresolved during that time period. Specific solutions in hardware, software, and personnel areseen as tactical questions, to be resolved in the light of current technology as they areencountered.

A single system image is proposed as the vehicle within which such tactical questions areresolved. Increased pluralism of native computing environments is seen for the future, with thesingle system image as the principal strategic element whereby coherency in computing andcommunications is maintained.

The image provides a single view of electronic mail, data base access, print and plot service,and archival storage for all users—whether on personal, departmental, or institutional comput-ing systems. The single system image offers a development prospectus that can be undertakenin small increments by highly disparate groups, allowing the total campus community toparticipate in its development. It may be viewed as completing the sixth and seventh layers ofthe ISO Reference Model.

A few additions and changes in terminology have been made in this current paper. I havecorrected a significant omission from the previous version of this work by including morediscussion of the role of the library in campus information systems and strategic planning. I amindebted to colleagues, too numerous to mention here, who have contributed significantly tothe formation of these perceptions.

Robert C. Heterick, Jr.May 1988

iv

Robert C. Heterick, Jr., is Vice President for Information Systems at Virginia Tech. Dr.Heterick has also served Virginia Tech as a professor of Management Science andDirector of the Design Science Center, as well as Director of the Computing Center. Hehas lectured widely on computer systems management and distributed computing innetworks of microprocessors, and served as a consultant on those areas to the IBMCorporation, the Governor of Virginia, the President's Privacy Protection StudyCommission, the National Academy of Sciences, and numerous universities and privatecorporations. He currently serves on the CAUSE Board of Directors.

About the Author

CAUSE appreciates the continued generous support of DigitalEquipment Corporation, who funded the publication of thisprofessional paper.

Acknowledgment

6/A SINGLE SYSTEM IMAGE: AN INFORMATION SYSTEMS STRATEGY

v

TABLE OF CONTENTS

1 AN INFORMATION SYSTEMS STRATEGY ------------------------------------------ 1

Adopting a Strategic View -------------------------------------------------------------- 1Central Computing IssuesPersonal Computing IssuesCommunications IssuesOffice Automation IssuesInstitutional Data Base Issues

Identifying a Strategic Course ---------------------------------------------------------- 4Communications NetworksComputing SupportProfessional Staff

Pluralism ------------------------------------------------------------------------------------ 7

Financing ----------------------------------------------------------------------------------- 7

2 A SINGLE SYSTEM IMAGE ------------------------------------------------------------- 9

The Image -------------------------------------------------------------------------------- 11Electronic MailArchive ServicePrint ServiceOffice Automation

Implementing the SSI ------------------------------------------------------------------ 13The CommunicatorThe AccessorThe DirectorThe Administrator

Issues for a Knowledge-Based Implementation ---------------------------------- 17The TranslatorHigh Performance Computing

3 A NEW PARADIGM -------------------------------------------------------------------- 20

An Information Systems Strategy/7

1AN INFORMATION SYSTEMS STRATEGY

Wherever we are, it is but a stage on the way to somewhere else, and whatever we do,however well we do it, it is only a preparation to do something else that shall be different.

` Robert Louis Stevenson

The growth of computing and communications technologyhas merged into a new area that we refer to as informationsystems—incorporating activities in computing, communi-cations, and data bases.

As we remap our strategic planning efforts, the time seemspropitious to ask if we are not at the point of another majorinstitutional commitment. It is generally agreed that theadvent of the digital computer signaled the end of the“industrial revolution” and the beginning of the “informationage.” As computing science and technology have maturedwe have begun to recognize that we may have overvaluedthe obvious computational capabilities of computing ma-chines and undervalued the informational capacities thatthey offer.

This new viewpoint suggests that it may be more than justsound management to rethink the institution’s commitmentto information systems technology. It may well be appropri-ate to set new directions and develop new initiatives in lightof emerging technology and changing economic parame-ters.

ADOPTING A STRATEGIC VIEW

Goals are important, but it is the strategies employed thatmatter the most.

Sun Tsu (3000 B.C.)

Classical approaches to planning usually emphasize theestablishment of goals. In a time where technology is grow-ing and and changing so rapidly, such a static approach isclearly myopic. What seems more fruitful is a strategic viewof the institution’s computing and communications future—a view that attempts to articulate a growth philosophy thatpermits seizing opportunities when the state of technology isright. Some technological advances are clearly predictable;others are not so easily foreseen. Whatever strategic position

Most institutions of higher education entered the world ofcomputing machines in the late 1950s or early 1960s withthe acquisition of first—or second—generation computers.Institutions continued to grow in their computing activitiesduring the ensuing five to ten years with a succession ofsecond-generation computers. During this time administra-tive and academic computing activities were established,generally as separate entities.

By the end of the 1960s most institutions made their secondmajor commitment to computing by moving into third-generation computers and, frequently, consolidating all oftheir computing activities under a single organizationalumbrella.

This second major commitment was accompanied by asignificant increase in the institution’s computing budget andappropriate changes in the institution’s organizational struc-ture to complement this new level of activity. From that time,institutional computing activities developed under the aus-pices of professional management—with the planning anddevelopment tools consistent with a major institutionalactivity.

We are now at another crossroad, marked technologically bythe maturation of the fourth generation of computing. Thelast several years have demonstrated implicit recognition ofthis by several occurrences:

• Many institutions have created a vice presidency for thechief information officer.

• A number of colleges and departments have begun torequire the purchase of personal computers by enteringstudents, and many others have the subject under activeconsideration.

• A new organizational structure to consolidate the com-munications activities (voice, video, and data) of theinstitution has been created or is under active study.


the institution assumes vis-a-vis computers and communica-tion, it must be predicated on foreseeable technologicaladvances, and flexible enough to accommodate those thatare not so easily discernible.

In attempting to develop a strategy that provides for theincorporation of new technology and the flexibility to adaptto unforeseen developments, it is useful first to outlineseveral philosophic opinions about the future of highereducation and computer and communications technology:

• Increasingly, institutions will perceive their constitu-ency to be off campus as well as on—the need to reachout across traditional boundaries will assume an in-creasing importance.

• Interconnection—the capability for human and ma-chine linkages—will become an important issue inhigher education.

• The fifth generation notwithstanding, the next round oftechnological innovation will occur in the communica-tions, not computer, arena.

• The capacity of the institutional budget to absorb con-tinuing developments in computers and communica-tions is significantly attenuated.

It is also useful to put into perspective the current institutionalcomputing environment by describing it in terms of threelevels: institutional, departmental, and personal. In manyrespects these levels may be equated to the “size” of com-puter: mainframe, minicomputer, and microcomputer.

Institutional computing has been with us the longest and hasbeen subjected to the most extensive study in terms ofdelivery economies, operational strategies, etc. This is theprimary base of computing on the campus, delivered gener-ally on mainframes, and accessed by users via time-sharingterminals. The bulk of computing power is concentratedhere, as is the predominant amount of electromagneticstorage and high quality printing and plotting equipment.

The primary source of departmental computing is special-purpose minicomputers. Where there is a relatively smallnumber (10 to 50) of individuals working on closely alliedproblems, it has been found to be cost effective to createspecialized computing environments.

Personal computing is a relatively new concept—one cre-ated by the confluence of VLSI and economies of scale inproduction and marketing. Personal computing permits thecreation of highly-specialized, custom-tailored computingenvironments for the individual. The relative newness ofpersonal computing may go a long way toward explainingwhy it is the least understood of the three computing environ-ments.

We currently have a somewhat mixed approach to supportof computing facilities. Institutional computing support is acentral administration task. The situation is less clear whendepartmental and personal computing support is consid-ered. Generally, departmental computing efforts may besupported centrally by a facilities management arrangement,or directly by the laboratory or department. There are excep-tions to this principle as epitomized by central ownershipand operation of minicomputers that belong in the depart-mental category. Personal computing is generally viewed asa college, departmental, or individual faculty or studentresponsibility. There are exceptions to this principle also, asevidenced by frequent computing center sponsorship ofpersonal computer laboratories.

A further step in developing a comprehensive strategy forinformation systems is to consider a number of questions thatcurrently present themselves:

• To what extent should computing requirements on cam-pus be satisfied centrally?

• What will be the role of the ubiquitous personal com-puter?

• To what extent should voice and data communicationsbe merged, and what are the roles of other subnetworkson campus?

• How should electronic mail and office automation besupported?

• Will access to the institutional data base become morewidely required, and are current data managementtechniques appropriate for the next five or ten years?

Each of these questions raises, in turn, a whole host of issuesand ancillary questions.

Central Computing Issues

We might begin by asking if central support of departmentaland personal computing is appropriate. Should institutionalcomputing be provided centrally, departmental computingby colleges and departments, and personal computing bydepartments, faculty, and students? A fundamental questionis to what extent can a central organization be expected tounderstand and respond to the diverse computing needs ofstudents, faculty, laboratories, and departments?

This raises the ancillary question of how computing fundsshould be distributed if some are to be allocated directly tocollege and departmental budgets. Clearly, financial supportfor computing is not unlimited. Just as clearly, the closer tothe user the funds are spent the better the user's computingneeds are likely to be satisfied (at least in the user’s percep-tion).

`


If computing needs are to be satisfied by a more decentral-ized philosophy, what percent of computing support shouldbe supplied by the institutional budget and what percent byindividual faculty, staff, and student purchase? Is it reason-able to expect students to acquire their own computingsupport? Or faculty and staff to acquire their own?

Finally, if a significant amount of computing resources issupplied by faculty, students, labs, departments, etc., whatimplications does this have for staff levels and skills in thecentral computing facility? Should individuals with com-puter skills be allocated directly to departments and collegesas is currently the case with many administrative depart-ments?

Personal Computing Issues

The time has come to assess the role of personal computingin higher education. An implicit assessment has been madeif one reads between the lines of computing policy changesover the last several years. Is it now time to make an explicitstatement of policy?

The cost of providing both ends of the communications linkfor terminals is quite high. The terminal/port ratio is quitelow. More widespread use of personal computers (and anoperational policy that supports this environment) can sig-nificantly improve the terminal/port ratio, thereby reducing(or shifting to other sources) the increasing cost of providingconnections to centrally supplied or supported computingand information resources. Strategies for supporting either anetwork of “dumb” terminals or a network of personalcomputers are significantly at variance. Is it time to adopt astrategy that favors connection of personal computers in thecampus network? Can personal computers provide the bulkof computational support required by the average studentand faculty member? Will they?

If personal computer requirements are not dictated centrally,and they probably shouldn’t be, how will the institutionavoid the chaos that might accompany undirected choice ofsystems? What should be the role of the central computingfacility in dealing with personal computing: should consult-ing activities be redirected toward the personal computer,should the central negotiation of hardware and softwarecontracts and site licenses for personal computer equipmentbe expanded?

Communications Issues

Communications issues are the major concerns of the nextfive years. Any strategy needs to consider to what extent, andhow, voice and data networks will be merged. To this mustalso be appended the question of video networks and theextent to which they should be merged with either voice and/or data.

Perhaps the predominant question is whether or not commu-nications networking should be a subsidized activity of theinstitution. That is, does the institution have an obligation toprovide communications access to staff, students and fac-ulty? Logically prior to that question is whether or not it iseconomically efficient or educationally necessary to providesuch interconnect capabilities? In what ways are the mis-sions of the institution enhanced by a more open set ofcommunication paths? Is a major communications network-ing capability critical to the growth of the institution? If theinstitution perceives a need or commitment to service off-campus constituencies, is a relatively wide area communica-tions network (voice, data, and video) necessary? important?

Are significant cost savings or avoidances available to theinstitution if it chooses to operate its own voice network? Areany potential cost savings offset by a new set of managerialor political problems? In the wide area domain, are thereopportunities for microwave operation or band sharing thatthe institution should investigate? Are there opportunities insatellite communication we should investigate, either forourselves or in concert with other agencies or institutions?

Will the local (or wide area) distribution of video (or com-bined voice, video, and data) see a significant increase?Should students have access to data networks from dormito-ries? Should they have access to video networks? If studentsin dormitories have such access, how will we handle stu-dents who live in town, or participate in off-campus instruc-tion? Should the institution provide faculty with access to thecampus data network from their homes?

Office Automation Issues

At most large institutions, current office automation andelectronic mail support are provided with institutional main-frame resources via terminal access. Will those users (pri-marily administrative) continue this mode of operation forthe next five years? Considering the number of terminalscurrently capable of connecting to the local area network,such “dumb” devices are probably in the minority andbecome an ever smaller minority every year. Should thecommunication network provide for intelligent device com-munication without passing through one of the networkhosts?

On most campuses the commitment to a mainframe textformatting system is significant and pervasive. Can personalcomputers provide an equivalent level of support? Will theproliferation of personal computers cause a significant up-surge in the number of large files passing through thecommunications network? Is there an office automationfacility available on personal computers to which we shouldencourage user migration?

Many campuses currently support many different electronicmail protocols, and several more are in use by individuals


and groups that are not supported centrally. Few of theseprotocols can be used to reach gateways outside the localcampus network. To what extent is off-campus electroniccommunication likely to be important in the future? Whatprotocol should the institution settle upon, and how shouldwe deal with the incompatibilities that this may cause withother heavily used software?

Institutional Data Base Issues

Many institutions made the decision to move to a data baseenvironment in the late 1960s or early 1970s. At that time,the most advanced transaction processing systems werehierarchical or network model systems which provide arelatively high level of security features and are generallyrobust. They are also expensive and old technology. For theaverage administrator on campus, spreadsheet relationalmodels are simpler to use and more consistent with whatthey do on personal computers. Is it appropriate to recon-sider the DBMS decision?

Relational technology has certainly matured, but is it suffi-ciently proven in its operational efficiency to consider mov-ing the institutional data base to a relational model? And if itis, which commercial product should we choose? Is itimportant to be able to download relational snapshots to database users? This could be done via current as well as by arelational system: Are the benefits of a relational systemsufficient to dictate a change?

In the more ubiquitous personal computer environment ofdown-loaded file snapshots, there is a host of operational,security, and privacy questions that do not have well-definedanswers. In the longer run, can we expect the institutionaldata base to be distributed across multiple CPUs? Should thehardware and software resources needed to support theinstitutional data base be considered a departmental-likecomputing resource? To what extent can user departmentsdevelop their own reporting systems if the institution onlymaintains a central data base and the associated data baseadministration responsibilities?

IDENTIFYING A STRATEGIC COURSE

It is a misfortune, inseparable from human affairs, that publicmeasures are rarely investigated with that spirit of modera-tion which is essential to a just estimate of their real tendencyto advance or obstruct the public good; and that this spirit ismore apt to be diminished than promoted by those occasionswhich require an unusual exercise of it.

James Madison The Federalist

The foregoing litany of questions doesn’t identify the strate-gic course we should set, but it does raise a sufficient set of

problems to suggest that it is appropriate to forge a strategicplan. Whatever strategy finally evolves, it should be one thatprovides a philosophy in which these and other questionscan be reconciled. In general, it is probably inappropriate totry to craft the strategy from specific answers to such ques-tions. Rather, we should attempt to discern what role theinstitution plans to establish for itself and, within the contextof how the institution views itself, set a compatible informa-tion systems strategy.

Underlying all the institution’s aspirations is the need for aneffective information systems infrastructure. This infrastruc-ture can be divided into three major components:

• Communication networks (voice, video, and data)

• Computing support (institutional, departmental, andpersonal)

• Professional staff (communications, computing, andmanagerial)

Communications Networks

Most institutions have only recently begun to recognize theeffort required to put into place a comprehensive communi-cations network. To a large extent, this is due to the autono-mous development of voice, video, and data communica-tions. The major step of placing all these activities under asingle management has yet to be taken for most institutions.

Voice service is probably the most broadly used of thecategories. The two principal planning issues in this arenahave to do with: (1) the merger of voice with video and datain the institution’s network; and (2) the associated questionof the appropriate division of responsibility between theinstitution and the local common carrier. For strategic pur-poses, it is probably useful to divide this issue on a geographi-cal basis, with those voice applications required off campusgenerally falling within the purview of a local operatingcompany, and those on campus that might be operated bythe institution.

It is not likely that the institution can provide voice servicelevels that exceed those provided by the common carrier.The rationale for institutional operation of voice servicesarises from the significant concern that common carrieroperation of the campus voice network leads to substantialsub-optimization. Put in the positive, by integrating voicewith existing video and data networks, there may be signifi-cant cost savings available through optimization of a largernetwork—avoidance of duplicative cabling, bandwidthsharing, etc. While a digital switching facility permits a morerational integration of voice and data services, it also presentsthe opportunity for new services. Principal among these isvoice mail, the ability to digitally store voice messages to bepicked up later by the intended recipient.


provide the principal base of computing support for theinstitution. If it will, then clearly we need to rethink ourapproach to institutional computing and set new directions.

One way to approach this issue might be to considerwhether, and under what circumstances, we can expect alarge percentage of students, faculty, and staff to haverelatively immediate access to personal computing. Cer-tainly by 1990 we can expect 32-bit, 20MHz (5-10 MIPS)processors that can be purchased at prices approximatelyequal to current-generation machines. This would imply thatthe power of today’s minis would be available in desktop(probably lapsize, if desired), personal machines. If theprices of computers continue their current trend, thentoday’s high-end machines will cost less than $1,000 in1990, and such machines should be powerful enough andsufficiently inexpensive that access to one could be taken forgranted. In brief, there seems to be no technological orfinancial impediment to assuming that by 1990, anyone whowants a personal computer will have one.

There are still a number of activities that should be providedcentrally as part of the institutional base. These wouldinclude the institutional data base, printing service, graphicsservice, archiving, supercomputing, and specialized soft-ware. One might very well extend the list, but the character-istics of what might make an institutional service item areprobably more important and interesting. An institutionalservice seems in order where a broad cross section of usersmust access the same data. This is certainly the case with theadministrative records of the institution and access to thelibrary catalogue.

An institutional service also seems in order for very high-priced peripheral devices. High-volume laser printing, highresolution graphics, microforms, photographic media, andtypesetting are typical devices that would fall into thiscategory. Another category of this type is high-volumestorage devices for machine-readable data.

There will continue to be a need for special purpose softwarewhich either must run on a large mainframe or is tooexpensive to acquire unless its cost is amortized over a largeruser community. Increasingly, we can expect a role forsupercomputing related to such software packages. Thesupercomputer of the 1990s will be on the order of 10 giga-FLOPS, bringing a new set of problems within the realm ofpractical computability.

If personal computing is expected to provide the bulk ofcomputing service for most users, then the institutionalservice should redirect its support efforts from “dumb”terminals to intelligent workstations. This transition will taketime, perhaps five years. During this transition period therewill likely be a number of temporary tasks that will alsobecome candidates for institutional support—store and for-

At the moment, and for the next several years, data commu-nication is the high growth area that is forcing decisionsregarding the communication infrastructure of the institu-tion. We can identify four major data communication areas:intraoffice networks; intrabuilding networks; interbuildingnetworks; and wide area networks.

The proliferation of personal computers is bringing with it ademand for intraoffice networking—the networking of sev-eral (perhaps up to six or eight) personal computers in adepartmental or laboratory office. Most users perceive thatthere will be a dominant node in this network, probablyproviding large fixed disk storage, high quality printing, orsome other sharable resource that is cost effective whendistributed across a half dozen or so devices. The advent of32-bit, time-shared micros will only accelerate this require-ment.

Such networks can be created with logic cards for thepersonal computers and cable that is installed departmen-tally. There exists the potential for literally hundreds of suchnetworks to develop—with a great diversity of products andprotocols. This is not necessarily bad, provided the networkcontains a server that provides a gateway to the campus (orintrabuilding) network. The difficulty with such homegrownnetworks is that they may prove to be unsatisfactory becauseof user naivete and consequently represent a less thandesirable expenditure of scarce resources.

While the development of intraoffice networks is debatablya central support issue, intrabuilding and interbuilding wir-ing and protocol standards must clearly be establishedcentrally. There are many issues here that require somedebate before standardization. As with most construction-related efforts, more cost-effective installation is likely iflonger-term needs and technological developments are con-sidered prior to initial installation.

The issue of wide-area networking is the most amorphous ofall, related as it is to the strategic plan of the institution interms of reaching off-campus constituencies. Clearly, themore serious the institution is in its plans to engage insignificant off-campus efforts, the more critical is the aspectof the infrastructure. Some rather specific guidance on thisissue is called for.

Computing Support

As defined previously, computing support can be looked atin terms of institutional, departmental, and personal effort. Inidentifying a strategy, it may be appropriate to start with thepersonal level, as this seems to be the current driving forcein the computing arena.

From the strategic viewpoint, the paramount question iswhether personal computing will, in the next five years,


ward of electronic mail, gateways to external networks, andfile transfer are examples.

The economics of computing technology suggests that de-partmental computing will become more important in thefuture. Our economy has always favored specialization, andvalue-added services in computing are becoming increas-ingly commonplace. If we accept the principle that depart-mental facilities service a limited base of users, we canidentify a number of computing services that would likely beof the departmental category: office automation, CAD/CAM,CAI and CMI, library automation, and robotics.

Again, it is probably more interesting to discern the charac-teristics that make a computing service a candidate for thedepartmental category. Clearly, one such characteristic isthe need for computing power beyond that likely to beavailable on the next several generations of personal com-puters. One characteristic of the preceding list is the inten-siveness of computing for the activities mentioned.

Perhaps the thorniest question regarding departmental com-puting is the source of funding. Where total funding is fromsoft sources such as grants and contracts this is not an issue.More generally, we need to resolve to what extent funding forsuch facilities should be considered an institutional ques-tion. If departmental facilities are at least partially subsidizedby the institution (and all are at least to the extent of certainoverhead costs), we need to determine what portion of thecost should be subsidized and how the institutional facilityplan for such acquisitions will provide for the disbursementof funds.

There is a second-order effect attributable to departmentalfacilities that may well raise issues more critical than those ofcost sharing. Departmental facilities require systems pro-gramming support, site preparation, gateways or bridginginto the campus communication network, and other laborintensive support activities. Replicating the skills available inthe institutional facility ten or more times over is probably notpossible, much less cost effective. To some extent, certain ofthese labor intensive skills will be provided by faculty andgraduate students. There is a real question as to whether thisis cost effective and whether the support level will beadequate.

Professional Staff

Some infrastructure questions relating to professional staffwere raised earlier with respect to central computing issues.The primary strategic focus in this area must be on skills thatwill be required, the number of personnel necessary, and thedistribution of those individuals. This concern is exacerbatedby the shortage of skilled computer professionals, the evengreater shortage of highly qualified technical managementpersonnel and, as a consequence, the extraordinary salarylevels commanded by individuals with these scarce talents.

Consider the skills likely to be of most value in the nextseveral years. As the campus builds toward thousands ofintelligent workstations, the first-level, and likely predomi-nant, need of users for information will concern their per-sonal computer hardware, software, and interface to variouscommunication networks. When we consider the number ofdifferent personal computers on the marketplace (probablyover one hundred) and the number of software packages(probably close to 10,000), we recognize that it is notpossible to provide expert consulting for all of them. Wemight possibly consider specializing in one or two machinesand perhaps a dozen pieces of software. Given that the half-life of micro hardware and software technology is about 18months, and given that at any time it is likely that threegenerations of the technology will be in common use oncampus, the problem of expert consulting still seems beyondreasonable grasp.

It seems inevitable that the burden of consulting will have toshift even further to faculty and students. Attempts to pro-scribe either hardware or software configurations seemantithetical to the purposes of a university and shouldprobably be avoided. The role of central computing staff inthe personal computer milieu needs to be addressed. Therewould seem to be several initiatives that will need to bemaintained centrally: standards for electronic mail, institu-tional printing access, institutional file service, system main-tenance of institutional software packages, and arrange-ments for quantity purchase of selected personal computerhardware and software.

The foregoing list is not all-inclusive, but is indicative of thetypes of activities that ought to be provided centrally. Thecommon characteristic of these activities is to support de-vices, protocols or software, rather than direct support of theuser. As the personal computer workstation becomes thenorm on campus, the capacity and capability to provide userconsulting is diminished, but the need to provide networkaccess to special services is amplified.

As the institutional service becomes more directed to devicesupport, it is reasonable to imagine direct user charges foritems produced, rather than for computing equipment exer-cised in their production. For instance, we can imaginecharges imposed for pages of output printed, slides ormicroforms produced, graphic frames plotted, etc. In thisenvironment we can envision page printing to be a service ofthe Print Shop rather than the Computing Center. We canenvision slide and graphic production to be a service of theLearning Resources Center rather than the Computing Cen-ter—the focus of institutional computing support shifting toprovision of the network intelligence to provide the intercon-nect capability for the devices in the network.


PLURALISM

There is nothing more difficult to carry out, nor moredoubtful of success, nor more dangerous to handle, than toinitiate a new order of things. For the reformer has enemiesin all who profit by the old order, and only luke-warmdefenders in all those who would profit by the new order.This luke-warmness arises partly from fear of their adversar-ies, who have the law in their favor; and partly from theincredulity of mankind, who do not truly believe in anythingnew until they have had actual experience of it.

Machiavelli The Prince (1513)

If there is one notable characteristic of this new generation ofcomputing it would have to be pluralism. The age of centrallydirected and supported computing has passed along withtyping pools, keypunch and word processing centers, andthe punched card itself. The economy of scale phenomenonapplies only to a few hardware devices attached to comput-ers and to the computer-communications infrastructure it-self.

Whatever strategic plan we craft must recognize this as theparamount fact of modern computing and communicationstechnology. The freedom to innovate is also the freedom tofail. As a consequence, we must be prepared to have groupspropose, and implement, computing environments that havelittle apparent chance of providing successful, long-termsolutions to the problems for which they have been devised.Hopefully, an occasional brilliant success that couldn’t havebeen predicted in advance will help offset some of the naivefailures.

Pluralism will be most evident in the personal and depart-mental computing areas. The tendency of users to wanthighly specialized computing environments that are mirrorimages of the latest technology being used in industry willaccelerate. Most of these environments will be clearlyephemeral, seldom having an industrial life of more than afew years. The truly difficult problem will be to encouragethe user community to find some way to teach principlesdevoid of their current technological trappings and packag-ing. As with any relatively new technology, it is difficult toascertain the underlying principles, and tough to argueagainst academic programs that want to provide “hands on”experience for students. The major brake on unnecessary orunwise technological acquisitions will have to come fromcarefully designed funding strategies.

In the personal computer arena, the tendency to want tomove almost annually to the latest technology will bedifficult to blunt. Aside from engineering and computerscience, it will be difficult to justify significant studentexpenditures for personal computers. An expenditure ofmore than a semester’s tuition on a computer is too much.

Most recent statistics suggest that the retail price of thesoftware being acquired for the personal computer is aboutequal to the cost of the machine itself. Requiring specifichardware or software configurations on the part of studentsin particular programs will eventually isolate those academicprograms from the institution at large. Unchecked, this kindof pluralism has the potential to destroy the institution as acommunity.

Along with the freedom to select various configurations ofsoftware and hardware must come an understanding of theneed to make generic computing requirements. When aspreadsheet is required it should not be specified to beprecisely Lotus, or VisiCalc, or ... . At this juncture the facultyhave done little to solve this problem, and until it is resolvedthe unchecked growth of personal computing must beviewed as a mixed blessing. Fundamental to our informationsystems strategy must be some mechanism to bring the issueto the table for consensual resolution. For anyone who iscurrently teaching courses in which the students frequentlyhave their own personal computers, the Tower of Babelanalogy is readily apparent.

This pluralistic computing environment will place increas-ing pressure on centrally provided services to have somecognizance of the multiplicity of software/hardware configu-rations that are attempting to access institutional services. Itis clearly not possible or economically sensible to centrallydesign software that is capable of printing a file edited on anyone of a hundred different what-you-see-is-what-you-gettext formatters. A similar observation could be made regard-ing downloading file snapshots in a format suitable for anyone of twenty commonly-used spreadsheet programs.

FINANCING

People who don't count, don't count.Anatole France

In at least one respect financing presents the most difficultaspect of crafting a strategic plan. It can be argued that it isnot the responsibility, or proper role, of Information Systemsto locate and develop funding sources—it should simplymanage, as effectively as possible, the funds allocated to it bythe institution. Of course, in the push and shove of thebudgetary process it can emerge with either more or lessfunding. The important aspect of the budgetary process isthat it won’t be much more, or much less.

However, if the perception that we are at the point ofconsidering another major institutional commitment to in-formation systems is correct, then funding cannot be ig-nored. For it is one thing to consider reallocating currentfunds and quite another to consider assuming initiativesbeyond those already assumed. If the strategy is built around


a relatively level (in constant dollars) budget, then questionsof what current services to restrict or curtail, what currentusers to reduce or eliminate services to, become paramountissues. If new funds are available, then service changes canbe phased in over longer periods of time, causing relativelyminor dislocations of user service levels.

It is probably appropriate to discern more subtle shifts offunding responsibility that have occurred in the last fewyears.

• Some colleges and departments require students toacquire their own computing resources.

• Many faculty have acquired their own personal comput-ers.

• Most colleges and departments provide access to per-sonal computers or terminals for their faculty and staff.

• A significant percentage of departmental grant funds areexpended on general purpose computing equipment.

• Colleges and departments have begun to assume thelocal area network connection cost in addition to con-nect time charges on mainframes.

All these charges, once covered out of the central computingbudget, have shifted to local units (colleges and departmentsprimarily), frequently with no commensurate shift of budg-eted computing funds. As a general rule, institutional com-puting service is a subsidized commodity, paid for centrallyin the computing budget, and rationed through a computerallocation scheme. Local and departmental computing re-sources are paid for from grant, college, or departmentalfunds accumulated through operating budget set-backs orexpenditure reallocations. Interconnect charges (modems,line costs, multiplexors, etc.) have risen from a negligibleitem in most departmental budgets several years ago to, formany departments, the single largest charge item in theirbudget. Capital expenditures for personal computers haverisen from almost nothing three years ago to, for manydepartments, the single largest capital outlay expenditure intheir budget. The institution, through equipment moderniza-tion and other supplemental programs, needs to contributesignificantly, although indirectly, to these acquisitions.

As a general rule, interconnect costs are direct costs paid bycolleges and departments from operating budgets. Commu-nication network developments are paced by revenues gen-erated from current services. This situation is tempered,although only slightly, by the previous investment in com-munications equipment paid for from subsidized computingoperations. It would take the wisdom of Solomon to untanglethis Gordian knot of support for computing and communica-tions. To some extent, the knot is severed with the establish-

ment of a consolidated communications group, but anunderlying rationale for funding disbursement seems not tohave been found—or at least not uniformly applied at mostinstitutions.

Funding is an extremely important issue because we seem tobe trying to continue doing all that we were doing before (andabsorbing natural and inflationary growth) and concurrentlystarting a major new program in voice, data, and videonetworking. The only way we can continue on this path andgrow the communications networks in the time frames andcost efficiencies that seem reasonable is with the infusion ofnew funds. Without, at this time, attempting a detailedassessment of funds required, and ignoring any shortfall offunding that may currently be perceived to exist, this shortfallrepresents the infusion of several million dollars on a one-time basis for most campuses.

While it is reasonable to believe that the largest part of thecommunications network costs can be shown to be recover-able in a five- to ten-year period (via more cost effectiveoperation by the institution as compared to a commoncarrier), such an argument is not one that should be made. Acost saving argument presupposes a “business as usual”condition that is significantly at odds with the perception thatthe institution is poised for a new spurt of activity whichinvolves significant changes in its goals trajectory. This nextset of aspirations of the institution requires the establishmentof a sophisticated computer and communications infrastruc-ture—the communications part of which is not in place, andthe computer part of which is in need of considerable tuning.

Institutions are increasingly turning to non-traditional fund-ing sources for needed infusions of capital and such a pathcould be pursued for the computer and communicationsinfrastructure. Without making any judgments here as to thedesirability or probability of success of alternate strategies,the following suggest themselves:

• Joint ventures with industrial partners in the computerand communications business;

• Major negotiation efforts with foundations that haveparticular interest in educational delivery systems, par-ticularly non-traditional systems;

• The trading of industrial park space with a suitableindustrial partner for a “named” project activity in com-puters and communications; and

• State- or institution-supported revenue bond issues.


2A SINGLE SYSTEM IMAGE

Software integration is the holy grail of information resource management. Daniel Appelton

The historic development of information systems hasbrought us through three readily definable stages:

• Transaction Processing

• Management Information

• Decision Support

The first, and immediate need, was for a system to supporthigh volume transaction processing in each of the functionalareas (accounting, purchasing, personnel, student records,etc.) of the institution. The data base management system(DBMS) needed to provide a high level of reliability (backupand restore features) and data item level security and authori-zation protection as well as high volume transaction support.

The second stage was reached when several of the functionalareas had been brought on-line and cross-area access couldbe used to produce management information. This manage-ment information was provided in the form of visual displayscreens providing pre-formatted responses to fixed queries,but on-demand, in user time (synchronously) rather than asscheduled, printed (asynchronous) reports. User requests forinformation were evaluated, and those with sufficiently highfrequency were programmed into the management informa-tion facilities of the system.

The system eventually would need to respond to ad hoc,unstructured queries coming from a much broader usercommunity. As a consequence of the proliferation of intelli-gent workstations, there has been a commensurate increasein distributed computing, native user environments (Lotus,BASIC, dBASE, etc.) and naive, first-time computer users.

This growth, and change in character, of the user communityhas brought, rather forcefully, to the attention of users thatthere is a significant difference in the requirements fortransaction processing and decision support. Not only are

there many more users desiring access to administrativeinformation, but this access is increasingly less structured—both in the form of information requests and the format inwhich the response is desired. This growth has fostered anequivalent growth in the backlog of management informa-tion system components—a backlog that is paralleledthroughout industry as well as higher education.

From the current vantage point, it appears that there are fourtypes of systems that are reasonably differentiable in theUniversity’s information systems environment:

TPS—Transaction Processing Systems play a fundamentalrole in the initial capture of data and provide the majorresource for functional offices.

MIS—Management Information Systems provide a broadcross section of secondary users with pre-formatted,easily accessible, query responses. These responsesfrequently require data from multiple functional areas.

DSS—Decision Support Systems are on the increase as usersbegin to develop significant computing capabilities ontheir own intelligent workstations. Unlike the MIS de-mand, this is primarily for access to information with theuser supplying the processing environment.

KBS—Knowledge-Based Systems appear to offer signifi-cantly enhanced decision support for naive users. Thesetypes of systems are extremely difficult to construct andare currently at the forefront of software engineeringresearch.

Viewed on the dimensions of “query structure” and “usercapability” (in the sense of computer competency) thesesystems appear as shown in Figure 1. TPSs are sometimesreferred to as “closed” systems, DSSs as “open” systems—theless structured the query, the more open a system is required.On the other dimension, the more naive the user, the lesscapable he or she will be of handling an open system.


There are several guiding principles of an institution’s infor-mation systems that are important and appropriate.

responsibilities include, at least, the availability of informa-tion that may be as unambiguously interpreted as possible.

This multi-level structure is frequently referred to as storage,conceptual, and user views of the data. Among the purposesof management information systems is the protection of theuser from needing to know the storage and, frequently, theconceptual schema; that is, making the user interface assimple as possible.

It seems natural that this process of making transparent andsimplifying interfaces be extended to the question earlierraised—should a single DBMS be chosen to address bothtransaction processing and decision support requirements?

In a properly segmented set of data views, the choice oftransaction processing system can be made, and changed,independently of decision support requirements.

When we consider the likely capabilities of workstations thatwill be available within five years, both in terms of process-ing capacity and data storage, such workstations will bemore than capable of supporting significantly large databases. Such machines are likely to be 20MHz, 32 bit deviceswith several MBytes of internal storage and hundreds ofMBytes of disk storage.1The availability of such workstationswill accelerate current trends toward both distributed dataand computing. In short, the administrative records of theinstitution are likely to be found, as well as used, in aheterogenous, distributed computing environment.

It also seems likely that access to many data bases, externalas well as internal, will be desired for decision supportpurposes. Such a view requires that we treat the issue ofadministrative systems as only a subset of a larger problem.When a solution to the larger problem is found, ease ofaccess to administrative data will naturally follow. It alsoseems likely that “dumb” terminals, while they will provideappropriate native computing environments for some, willbecome a distinct minority in the mix of workstations oncampus.

The critical issue for users with intelligent workstations willbe access, not processing. While there will continue to exista class of user who desires some pre-defined processing ofinformation from a data base (what we have referred to as theMIS function), the vast majority of users will wish to do theirown processing. Users who desire pre-defined processingwill continue to depend on some other organization tosupport their interface. As a general principle, it seemspreferable for this support function to be located as close tothe user (and the data custodian) as possible. This suggests

• Source point data capture

• Value-added data handling

• Destination point documentation

These are principles associated with the transaction process-ing aspect of the institution’s information systems. Nothing inany perception of the changing information systems environ-ment on campus suggests that transaction processing is notstill an important component.

What is equally evident, and superficially paradoxical, is thatthe most appropriate decision regarding transaction process-ing may well be most inappropriate when viewed from thedomain of decision support. At least two courses of actionsuggest themselves:

• Find a DBMS that strikes the best balance between theapparent conflicts of the two needs.

• Find a way to separate the two activities so that choicesfor data base management and decision support may bemade in restricted, non-overlapping domains.

It is the data custodian who must define in what way, howoften, and by whom data may be captured, updated, ac-cessed and deleted. In the process of this definition there willarise, at least implicitly, information upon which decisionsregarding DBMSs may be made. The custodian might welldesire data to be collected and stored at one level of detail,but accessed for decision purposes at another. In fact, acompelling case can be made that the typical decision makermight not even be able to navigate through the data diction-ary for the data kept at its most primitive level. The custodianmay well decide that there are some data not available fordirect query by users. The custodian has responsibilitiesbeyond the simple capture and storage of information. Those

1 John P. Crecine, “The Next Generation of Personal Com-puters,” Science, 28 February 1986, pp. 935-943, containssome prognostications of interest.

Figure 1. Open-Closed System Spectrum

high TPS MIS

low KBS DSS

naive skilled

User Capability

QueryStructure


that an increasing amount of support will come from func-tional offices (particularly the custodian’s office) rather thanfrom a systems development group.

These intermediaries (custodian and functional offices) willhave the same characteristics as the bulk of the user popula-tion with intelligent workstations. Their major concern willalso be ease of access. In addition, the custodian must defineand share security responsibilities. In the distributed environ-ment privacy will be considerably more difficult to insure.While it is true that the major change has come in the domainin which security is to be invoked—from paper, to central-ized (primarily hard-wired) terminal systems, and now todistributed data and processing—the threshold of improperaccess may both be lower and not seem to require blatantlyillegal action on the part of the trespasser. At least, a newround of user education is called for.

THE IMAGE

... a distributed system should look like a centralized systemto the user. The problems of distributed systems are (orshould be) system problems, not user problems.

C. J. Date

The proliferation of personal computers into the institutionhas occurred so rapidly that the central computing andcommunications organizations have not had the time orresources to adequately provide an appropriate computingenvironment for them. To date, most institutions have sup-ported terminal emulation, aided by file transfer protocols, asthe only means for the personal computer user to access thecentral organization’s data base and peripherals. Electronicmail is usually accessed via terminal emulation on one of theinstitution’s mainframe computing systems.

This situation raises the threshold of detailed knowledgerequired of the user—precisely the opposite of what thesituation should be and the inverse of the reason most usersacquired a personal computer in the first place. The lack ofappropriate networking support has only exacerbated thesituation.

The situation is similar for users of special purpose minicom-puter systems who have chosen such a system for their nativeenvironment in order to have access to CAD/CAM, labora-tory automation, specialized operating systems, etc. In fact,many mainframe users can be similarly classified. Someusers have chosen a mainframe environment for access tosophisticated office automation facilities or access to spe-cialized compilers or software not available on the currentcrop of personal computers.

We can use the term “native environment” to encompass thisbroad category of users who have selected a micro, mini, or

mainframe subsystem for what it has to offer their individualcomputational and information needs. Even when registeredon one of the host mainframes, these users may have anextremely limited knowledge of their complete environ-ment, choosing to see themselves as office automation usersfor instance, not knowledgeable in the whole gamut of otheroperating system services available to them. Further, they areprobably not the least bit interested in learning the protocolsnecessary to use them.

What these users need and want is an institutional environ-ment in which they can use the native environment for thespecific set of tasks for which it was selected, withoutforfeiting access to electronic mail, organizational databases, or sophisticated peripherals found in theorganization’s computing network. Accompanying the pro-liferation of personal computers has been a commensurateproliferation of user native environments—Lotus, dBASE,Wordstar, BASIC, etc. The user needs a simple extension tothe native environment that provides access to the aforemen-tioned organizational resources. Further, that extensionshould appear the same to every user, irrespective of theparticular native environment.

We can refer to this extension as a single system image (SSI).The user wishes to see the image of a single system—onewhich is a natural extension of his or her native environment.Such an image should be free of specific protocols associatedwith one or more of the institution’s central computingsystems. In fact, any association with one of those systemsshould be totally transparent.

It would appear that the future of intelligent workstationsversus shared-resource, interactive computing will parallelthat of the private auto and mass transit. While the public hasindicated that they believe in the economies of mass transit,they generally find it individually dysfunctional—or at leastpersonally restrictive—to the point that they continue toprefer the private auto. Much of the same psychology entersinto the decision of personal versus shared-resource comput-ing. The first wave of the personal computing frenzy mayhave begun to recede, but as ever more sophisticated work-stations come to the market, we can expect personal comput-ing to proliferate even further.

The image could be implemented as an extended electronicmail system. In addition to a conventional mail facility thatprovided access to a half dozen or more national networks(BITNET, ARPAnet, CSNET, NSFNET, etc.), the mail systemcould provide access to organizational data bases, highspeed laser printers, typesetting equipment, flat bed plotters,high capacity disk drives, etc. The plethora of native environ-ments suggests that the single system image will be complexand its development will need to be shared between thecentral computing and communications organizations andthe community of personal computer users.


Synchronous access to distributed data is still at the frontierof research in computer science. To be sure, there are severaldata base systems that purport to manage distributed data. Inevery case a similar software package resides on a similarhardware architecture. This is not the problem that manyinstitutions will need to address. Commercially viable ver-sions of a solution are not even likely in the next five years.On the other hand, an asynchronous solution such as elec-tronic mail appears to offer most of the functionality of thesolution we would hope to see—and it appears that it can beconstructed in a relatively short time with limited resources.

Electronic Mail

A mail service would need to permit the user to compose themail item in his or her native environment and communicatethe item to the network and from there via network servicemachines and/or host mainframe facilities to the intendedrecipient. The actual transfer of the mail item could be via apersonal computer file to a host file, but should not berestricted to such. In fact, many, if not most, personalcomputer users will opt to transfer an object which is not aDOS or UNIX file. The object might be one or more cells froma spreadsheet, a programming language variable, or anobject created by some piece of “desk organizer” software.

From the native environment, the user would need to con-nect to the network. Having made such a connection, amessage to a host or service machine would invoke someexecutive routine which would accept the mail item from thepersonal computer (this embodies some or all of the charac-teristics of a file transfer protocol such as Kermit orPCTRANS) and dispatch it via the host system or servicemachine electronic mail facilities. The user need know onlythe name of the executive routine to invoke and the user ID,node, and network of the intended recipient.

Many current workstations are limited to 19.2 Kbps and areconfigured with an inexpensive asynchronous interface. Asmany users will access the system in a dial-up mode, suchasynchronous connectivity must be supported. However, toget maximum efficiency from the network and to avoiddedicating workstations during large file transfers, a highefficiency protocol should be provided for.2 The next genera-tion of workstations will likely be true multitasking machineswith higher I/O port transfer rates.

The user should have the facility to request that a particularhost or server system file be sent to the personal computer.The mail item would be displayed on the screen, captured bya native environment variable, or placed in a workstationfile. Some users may prefer to archive mail items on a

network server or host system, others will prefer saving mailitems in a workstation file or as some type of object in theirnative environment.

Archive Service

The personal computer user would likely desire some formof archive service from the network. An archive wouldprovide for disposition of mail histories, workstation filebackups, or temporary high volume storage external to thepersonal computer. The economics of storage technologyindicate that large volume storage on network hosts orservice machines will continue to be the most inexpensiveapproach until multiple write laser disc technology is com-mercially available.

The archive could be implemented via electronic mail bymeans of an executive routine capable of storing and retriev-ing items using host system or service machine protocols.The user could mail an item to the archive for storage or maila request to the archive to retrieve a previously stored item.

Print Service

A rudimentary print service could be quite easily imple-mented as an electronic mail facility. The major stumblingblocks involve the location (personal computer or host)where the file was formatted and the compatibility of theescape sequences for print control in the formatted objectand the target printing device. For many of the what-you-see-is-what-you-get personal computer formatters this is far froma trivial problem.

In order to accommodate both formatting strategies it isprobably appropriate to devise an institutional printing stan-dard for word processing codes which would permit anyformatted file to be “sifted” and altered to the standard.Translation from the standard to specific network devicescould then be constructed (there should not be many of themneeded) to map the file to a specific printing device.

Most special printers will require that the font combinationsdesired also be specified—probably in the request ratherthan in the file to be printed. There are also potentialproblems associated with specifying the type and/or size ofthe paper to be used. Neither of these issues appears topresent significant problems that could not be easily accom-modated in a mail format.

The definition of a standard set of word processing codes willbe a difficult problem—one sure to raise heated debate fromseveral quarters. There are several emerging “standards” forboth layout and mark-up. A cursory review of them suggeststhat they will not accomplish what is required of the standardproposed here, although they may offer many good pointersfor such a specification.

2 See, for instance, Andrew Tannenbaum, Computer Net-works (New York, NY: Prentice-Hall,1988), for additionaldiscussion of sliding window protocols.


Office Automation

We can expect many users will, for some time, continue toprefer an office automation environment as a terminal useron a central computing facility. These users will desire tohave the processing routines and interface to the singlesystem image constructed for them. There seems little advan-tage in constructing similar interfaces which would operateon intelligent workstations. When data base access is addedto the functionality already present in such systems we mightexpect a significant new demand for services to develop.

IMPLEMENTING THE SSI

Nothing will ever be accomplished if all possible objectionsmust be first overcome.

Samuel Johnson

Access to distributed data bases is considerably complicatedby questions of authorization, privacy, security, etc. Ignor-ing, for the moment, these very real and important questions,we could formulate a simple electronic mail environment forgaining access to a data base. It seems fair to note that thisenvironment would dramatically affect both the number andsize of messages flowing through the campus communica-tion networks.

Access could be specified in terms of some real or somecanonical language. For our purposes here we will take theStructured Query Language (SQL) as a model. SQL, as mostother data base languages, has operators for file creation,record addition, modification and deletion, and generally anumber of other useful functions. We will temporarily re-strict our consideration to record selection from a relationalmodel of the data. We will assume that some set of routinesis capable of producing a relational view, irrespective of theactual data model implemented in the data base and the database management system. Further, we will assume that somemetadata definitions of actual or logical relations (userviews) have been implemented to relieve the user of thenecessity of having detailed knowledge of the primitive database design and data dictionary.

Conceptually, access to data requires navigating throughboth the single system image and the DBMS of the target data.The DBMS is required to physically manipulate the dataelements of each particular data base (although there may beparticipating “data bases” that are simple flat or sequentialfiles with no DBMS) and there may be a large number ofDBMSs involved. The single system image then is data aboutdata (metadata) that provides information about the physicallocation of data bases, rules to decide if some canonicallanguage request is properly formed, first-level security andauthorization checks and translation rules.

Figure 2 shows a summary of the roles of application routinesand the DBMS or SSI for different data types that may befound in the data bases participating in the SSI.

Figure 2. SSI Conceptual Model

DATA TYPE

Numeric

Text

Geometric

Image(bitmaps)

Application

arithmeticsortetc.

catenatesubstringetc.

intersectunionetc.

clip composite

DBMS

store

retrieve

update

etc.

Operation

logicalrelational

lexical

relations

SSI

DIRECTOR

ADMINIS-TRATOR

ACCESSOR

TRANSLA-TOR

DATABASE METADATA

A conceptual model of the image could be constructed of fivefunctions:

COMMUNICATOR—the hardware, software and protocolsrequired to physically transport data between source andtarget systems.

ACCESSOR—the data manipulation functions required toquery (perhaps store) data anywhere in the single systemimage, and to route data (perhaps electronic mail, printfiles, etc.) to a target.

TRANSLATOR—the ability to translate data and/or ACCES-SOR commands from source to target formats.

DIRECTOR—knowledge of where data may be found or howtargets may be accessed, if data need to be translated orheaders modified, how the target is addressed, etc.

ADMINISTRATOR—rules for access and update authority,invocation of consistency requirements, privacy andintegrity requirements, etc.

We can consider the COMMUNICATOR role to be played bythe local campus networks and discuss only some low-levelprotocol issues. The ACCESSOR, DIRECTOR, and ADMIN-ISTRATOR functions of the image, however, must be definedbeyond a brief discussion of low-level protocols. Finally, therole of the TRANSLATOR is deferred to the discussion of aknowledge-based representation of the image in the nextsection.


The COMMUNICATOR

At least initially, most of the workstations participating in thesingle system image would communicate with a servicemachine using existing asynchronous interfaces and net-works. Asynchronous serial communication has tradition-ally been used for communication of dumb terminals withhost computer systems. In this environment, data are typi-cally transmitted in an unstructured format. Such transmis-sion is not appropriate for the reliable exchange of largeamounts of data between computer systems in the singlesystem image. Electrical noise and marginal electronic com-ponents can cause transmitted data to be lost or corrupted.

A low-level protocol can be implemented on an asynchro-nous network to insure data integrity and prevent loss of data.Use of block sequence numbers, error detection codes, andretransmission of corrupted or lost blocks are the usualtechniques for constructing protocols that ensure data integ-rity. In addition to insuring reliable data transmission, theprotocol used for the single system image should be efficientin order to allow large amounts of data to be transmitted overnetworks that function at relatively low data rates. Protocolsthat take advantage of the full-duplex nature of modernasynchronous networks are prime candidates for the SSI low-level protocol. Sliding window and piggybacking protocolsallow the sender to transmit several blocks before receivingan acknowledgement for the first block, thus significantlyreducing the wait time between transmissions.

The protocol used for the single system image could be usedto implement a limited security scheme. Messages ex-changed by a workstation and network service machinecould include a session password. This password could beassigned by the network server during an initial identificationsequence during which time the user is validated by a userpassword. Following the initial sequence, all messages ex-changed by the workstation and the network service ma-chine would contain the session password. Since a worksta-tion may connect to and disconnect from the network manytimes during an active session, the session password wouldallow a session to be identified without the need to invoke thefull user validation exchange during each reconnection. Thesession password would also prevent session confusion if theconnection between the service machine and a workstationwere prematurely terminated, and a second workstationbecame substituted for the first. For security reasons, thesession password should only be stored in volatile RAMmemory.

The low-level protocol should be implemented on all work-stations that communicate on unstructured asynchronouslines. (Some computers participating in the single systemimage, particularly hosts, might communicate using other,commercially available, reliable protocols.) The initialimplementation would require protocol drivers on the net-work service machine and popular workstations.

One question of concern is network delay. That is, to whatextent would naturally-occurring delays in the networkcause a mail-based system to operate beyond the bounds ofconvenient user time? For our preliminary purposes here wemight think of network response time of a few seconds fortrivial actions (e.g. confirmation that a mail item has beenrouted to its destination locally) to a minute or so forresponses to significant actions (e.g. return mail of a signifi-cant data base query). At this juncture we are not concernedwith delays imparted by the software in handling the action,but communication network delays in actually passing theresults.

Using a full duplex, asynchronous protocol for mail activitieswithin the image (intelligent workstation to network server),there would be three major areas where delays might occur:

NETWORK—the effective communication speed (or band-width) of the underlying local area networks.

PROCESSOR—the effective processing speeds of the work-station and network servers.

DISK I/O—the effective data transport rates between CPUand disk on the workstations and network servers.

If the server and workstation were very fast (and they will bewithin the very near future), then it is clear that the networkdelay would be the major bottleneck for typical “trivial“queries. Network delay, of course, would vary with theinterface to the network being utilized; speeds of up to 64 Kbsare possible on PBX systems, with much lower speeds ontypical dial-up modems. However, even 64 Kbs will not besufficient as large data queries and file transfers becomecommonplace. High-speed networking alternatives, such asEthernet and Token Ring, supporting megabit speeds, willneed to be investigated.

The ACCESSOR

The role of the ACCESSOR would be to provide a canonicallanguage which permits and facilitates access to compo-nents of the image.3 The canonical language need notactually be embodied in code, but rather would act as aninterface standard which everyone would write to. The user,desiring elements from one or more data bases, wouldcompose his or her request in the canonical language. Thevarious data base administrators would provide “front ends”to their DBMSs which would translate canonical languagerequests to well formed calls in the language of the DBMS.This process would make the storage and conceptual viewsof each participating data base transparent and permit alldata bases to be addressed with the same command struc-ture.

3 The word “neutral” is frequently used in place of “canoni-cal,” particularly in the context of IGES and MAP.


For the naive user, the interface to the canonical languagemight be mediated in several additional levels. For instance,an office automation system might call up Query-by-Ex-ample-type screens to provide an iconic interface to thecanonical language. Similar types of “front-ends” will likelybe developed by personal computer users for the morepopular native environments such as Lotus, dBASE andBASIC. Minicomputer system administrators might well de-cide that similar facilities should be included in the operatingenvironment for their systems.

The choice of canonical language should not be madewithout extensive study. In the image concept, DBMSs maycome and go, but the canonical language used for accesslives on forever. The canonical language might be selectedfrom one of the popular fourth generation languages such asFOCUS or NOMAD. However, it is not clear that suchproducts will ever garner sufficient market share to becomede facto standards. It seem more likely that de jure standardswill evolve from national standards sources. If we had toguess what might emerge, we would go with SQL.

If an existing definition such as SQL is chosen, it will still benecessary to consider non-standard extensions to provide formail, archiving, printer access, etc. which are not part of theconceptual model language implementations.

The definition of user views falls to the purview of the datacustodians. In the case of complex data bases such as thoserepresented currently in most administrative systems, thecustodian will need to define a set of data fields that is notnecessarily a subset of those actually maintained in the database. For instance, it may require a dozen fields and hun-dreds of computations to produce a normalized annualsalary—the data element that the custodian deems mostappropriate for query purposes.

In addition to defining the logical fields of the user view, thecustodian will need to define a set of access rules—authori-zation to view all or a subset of the fields and all or a subsetof the records. For many data bases this will present minimalproblems as the entire data base will compose the user viewand there will be no restrictions on fields or records. For theinstitution’s administrative files this is clearly not the case.The “front-end” that prefaces access to this data base willneed a great deal of thought. This general subject is surveyedmore completely in regard to the ADMINISTRATOR.

For some data bases, particularly those administrative databases of the institution, it should be possible to bypass theimage altogether and access the data base directly withfourth generation products such as FOCUS or NOMAD. Thismay eventually be the access of choice for the functionaloffices and groups who are building MIS-like interfaces toone or more of these data bases.

The DIRECTOR

The DIRECTOR task can be envisioned as sharing responsi-bilities with the data dictionaries of the participating databases. A user request, formulated in the canonical language,would first go to the DIRECTOR, who would have to parsethe request to decide several issues:

• Is the request well formed syntactically?

• Are the target data bases referenced known to theDirector?

• Are data paths and data formats defined for the target(s)?

• Is more than one target referenced in the request?

• Are there institutional administrative tasks that must beperformed—first level security or authorization checks?

• Etcetera

The list is summarily terminated with an “etcetera” becausethese are typical of the issues that a system design wouldaddress. I will not attempt a design here, but rather willidentify what seem to be the thorniest problems and theirpotential solutions. The goal will be to suggest that the designproblems can be solved at reasonable cost, in reasonabletime, with current technology.

To be maximally useful, the image should be able to handlea request that references more than one data source. Themajor problem here is to decide whether such requestswould be serviced synchronously or asynchronously. Ifasynchronous responses are acceptable, the user wouldreceive two or more responses to a request, one for eachtarget accessed. The user would then have the responsibilityof merging the responses to make a composite response.While this might be acceptable in many cases, it wouldprohibit cross data base queries where one data base must befirst accessed to develop the access criteria for a second database.

If the decision is for asynchronous treatment of compositequeries, it should be seen as a short-term, temporary solu-tion. More thought needs to be given to the frequency andnecessity of sequential routing through multiple data bases.

Even if sequential routing is not required, the asynchronousarrival of responses to a composite request is still less thansatisfactory—too much of the burden is allowed to rest withthe user. A synchronous response to a composite requestwould require that the DIRECTOR assume many of thecharacteristics of a DBMS itself, in fact, to display somecharacteristics that are not currently found in DBMSs. TheDIRECTOR would need to have a memory and a temporary


storage, and be able to hold partial responses until all havecleared, at which time it would recompose a compositeresponse to the user.

There is also the question of how much about the data basestructures should be known to the DIRECTOR. Should theDIRECTOR have knowledge of the domains and/or fields ofthe participating data bases? If it did, it could do additionalcontent checking of requests, over and above verifyingtargets. This would permit a reduction in network traffic,which would be exceptional in a distributed environment. Itseems that a much “cleaner” set of interfaces would emergefrom a design that does not attempt to do anticipatory contentchecking—one that checks content only when levels arereached where content is already defined.

It may be appropriate at this point to observe that not all thedata sources participating in the image need be on campus,or even under the control of institutional employees. It is easyto imagine a pseudo data base that mimics some commer-cially available data base. The role of the pseudo data baseis to be the “front-end” for the commercial data base. In suchcases, the “front-end” would not be the responsibility of thedata administrator for that data base, but some intermediaryon campus who assumes this surrogate role.

The DIRECTOR would also need to be able to communicatewith the ADMINISTRATOR when security or authorizationchecks are required at the image level. Such communicationshould probably be synchronous. If it is, the prospect ofdeadlocks would need to be contended with.

DIRECTOR services would also require some additions tothe canonical language—at least in the sense of an errormessage methodology. A minimal set of error messageswould include:

SYNTAX—the original (and any derivative) request returnedwith some indication of the point in the request where itbecame syntactically unparsable.

VALUE—the original (and any derivative) request returnedwith the offending name.

DOMAIN—the original (and any derivative) request re-turned with the name of the offending operator wheredata were not in the domain of the operation requested.

These messages should be constructed so that they could bereceived and operated upon by an automated process.

Handling update requests would place a number of addi-tional burdens on the DIRECTOR, among them questions ofrecovery and what to do with composite updates when oneof the data bases is not “up” (a form of the concurrencyproblem). It is certainly possible to update files and data

bases asynchronously via electronic mail. Major difficultiesmight arise relative to edit checks and decisions as towhether to do partial updates when some of the data fail topass an edit check. It is also likely the case that many updateswould most naturally be expressed in a format other than thatsupported by the canonical language. This would require thedevelopment (or modification) of a significant amount ofsoftware to convert these more natural formats (e.g., report-ing class grades directly from faculty) into the canonicalform. The update problem may indicate the need for someshared file capability within the image functions.

The DIRECTOR would need some means of communicatingwith users to allow them to inform themselves of institutionalservices available in the image. There would also likely needto be some form of HELP services for syntax of the ACCES-SOR. There is a distinctly difficult question about whether theDIRECTOR should know about fields that exist in the userviews of the participating data bases. If it does, and is capableof responding to user requests about information at the fieldlevel which is available, it would likely also have to knowabout authorization rules for those fields. This might be seenas an ADMINISTRATOR question, but the real problem hasto do with the distribution of authorization security betweenthe image and the participating DBMSs. Further considera-tion of this question is provided in the section dealing withthe ADMINISTRATOR.

The DIRECTOR is most easily envisioned as a single “ma-chine” maintaining its own directly accessed storage devices(DASD). The DIRECTOR would need the support of gateway“machines” to facilitate access to other networks or to usersattached to multi-access machines. It is likely that a singleDIRECTOR “machine” would experience significant per-formance degradation when there were any substantialnumber of image users. It might well be appropriate topartition the DIRECTOR services functionally among a maildirector, a print director and an archive director. This mighthave the unwelcome effect of requiring users to know towhich of the directors a specific task should be directed. Analternate strategy might be to provide a shared file systemamong the directors.

The ADMINISTRATOR

The ADMINISTRATOR function presents the most difficultyin the implementation of the image. Authorization andsecurity are presumed to be ADMINISTRATOR functions.There is an overriding question of network security in generalwhich, while not directly an ADMINISTRATOR function,might best be discussed at this point.The image would necessarily increase significantly the vol-ume of traffic flowing through the institution’s communica-tion networks. If we ignore the need for physical and electro-magnetic radiation security of the actual physical links(something we might not be able to do in the long run) there


still exist the network security questions related to whetherthis message is actually from whom the packet envelope saysit is, and whether its contents have been altered. These arenon-trivial questions without obvious answers. At least ini-tially, something such as a session password would providea level of security commensurate with that currently avail-able in most campus networks. In the longer term it would benecessary to add data encryption to the network.

The only authorization checking that should be done by theADMINISTRATOR is at the global level. That is, if theDIRECTOR is limited to knowing only of the existence of databases and nothing at a lower level (fields and domains), thenit is reasonable to have the ADMINISTRATOR know only ofauthorization at that same level. For instance, is this user (orrepresentative of a particular user class) permitted to know ofthe existence of a certain data base? If he is, then furtherrequests for field-level information would be passed on to thefront end of the requested data base; if not the request wouldbe terminated with some appropriate response.4

Having moved nearly all the difficult questions from thedomain of the image to the front ends of the participating databases, it is probably appropriate to consider the developmentproblems engendered.

Most of our ADMINISTRATOR discussion has taken place inthe context of either DBMS front ends or data base directormachines. It may not prove useful to continue to try toseparate the administrative from directory functions. We willdo so here on the principle that maximum segregation offunction would make the image easier to maintain andenhance. Automated data base administration would servetwo functions:

DICTIONARY—the ADMINISTRATOR should provideminimal data dictionary functions for those data servermachines that otherwise would not have a data diction-ary. For data servers with a formal DBMS it would needto point to the appropriate data dictionary, avoiding theduplication of that information.

AUTHORIZATION—it would need to handle authorizationfor those data server machines that did not have anauthorization facility. Similarly, for data servers with aformal DBMS it would point to that facility.

ISSUES FOR A KNOWLEDGE-BASEDIMPLEMENTATION

Any sufficiently advanced technology is indistinguishablefrom magic.

Arthur Clarke Profiles of the Future

Ultimately, the image evolves into a major knowledge-basedsystem providing a standard specification of all systemsparticipating in the single system image. It incorporates someof the functions of data administration, user services andnetwork services into an “intelligent” communication inter-face.

As a strategy, the single system image provides a develop-ment prospectus that can be undertaken in very smallincrements by highly disparate groups. It has utility at anystage of its development; the more fully developed, the moreutility. Further, it is transparent to the user, requiring the userto know only the ACCESSOR commands in addition to themanipulation requirements of his or her own native environ-ment. Ultimately, it makes the entire computer and commu-nications infrastructure transparent to the user, permittingdata maintainers to be concerned primarily with the collec-tion and storage of data and only secondarily with useraccess to it. Hardware, software and communications tech-nology may be changed without impacting users. In thiscontext, the image can be seen to be a transparent interfacebetween the user’s native environment and the data storageand manipulation environment of the target system (anotheruser’s mail files, a system printing device, an IMS data base,etc.).

There are a number of design issues that could be dealt withmore or less independently, suggesting that the project couldbe fast-tracked. Among these issues are:

• Design of the low level, high efficiency, communica-tions protocol

• Design of a standard electronic mail format

• Design of gateway “machines” for off-campus elec-tronic mail

• Design of a “neutral” text formatting standard

• Design of a shared file archival service

• Design of a security system, perhaps including dataencryption

Perhaps the greatest advantage of the single system imagewould be the necessity to publish formal documentation andinterface standards. Perhaps the best way to disseminate thesystem standards would be via a bulletin board service. This

4 This might be grounds for adding another “error” messageto our short list.


would permit users to contribute their own interface code foruse by others. The quality of such unsolicited code wouldlikely be quite variable (witness the code for various imple-mentations of Kermit maintained by Columbia University). Itmight be desirable and/or necessary to enforce some sort ofquality control over user-submitted code but, even so, itwould have the advantage of permitting widespread partici-pation in the development of the image.

The TRANSLATOR

The long-term goal for the TRANSLATOR (See SSI Concep-tual Model, Figure 2, page 13) should be to support the user’snative computing environment by permitting the user totransmit and receive in data formats of that environment.Because there would likely be a very large number of nativeenvironments the TRANSLATOR will likely evolve into aknowledge-based system.

In the early phases of the image it should be possible tosupport several of the more broadly used data formats. Forinstance, several DOS-based environments (SuperCalc anddBASE) support data transport using Comma Separated For-mat (CSF) ASCII files. Many popular packages can generateCSF files from their internal storage schema and can read CSFfiles and reformat them into their internal storage schema. Itwould be useful to create TRANSLATOR functions to convertcolumnar data into CSF. Such functions should be relativelysimple as all that is required is to quote strings and separatefields with commas. Headings and titles, particularly inspreadsheets, would significantly complicate the task. Theinverse function of creating a columnar format from CSFwould also be useful. One obvious use of such functionswould be to put a “flat file” obtained from a data base queryinto a spreadsheet.

This type of translation is only adequate when the computedcell values of a spreadsheet or the data items from a data baseare required. Such a translation would not preserve thestructure of the data for data bases (e.g., the size and type offields). Neither would it preserve the actual contents of cellsthat contain computational formulas. In the early phase ofthe image some relatively simple TRANSLATOR functionscould be produced to aid in translation of the most popularspreadsheets and data bases. These functions would bebased upon low-level code, would not be general, andwould require maintenance as versions of the softwarechanged.

There is an important issue relative to the location of theTRANSLATOR functions. They might either reside on anetwork service machine, or be distributed as workstationprograms. Maintaining the functions on a server has theadvantage of single-point maintenance and update with onlyone version for each translation. The major disadvantagewould be the additional time required to transmit files to andfrom the server.

Ultimately, the image would require a knowledge-basedimplementation that would promulgate a “data descriptionlanguage.” At the lowest level, the internal representation ofprimitive data elements would need to be described. TheTRANSLATOR would need to know that a particular nativeenvironment spreadsheet stored floating-point data in 32-bit, excess-64, normalized format with a hidden leadingmantissa bit. A particular description for strings might statethat characters are stored in 8-bit bytes using 7-bit ASCII withhigh order bits cleared and a NULL delimiting the end ofstring. The TRANSLATOR would need to contain descrip-tions for most of the popular file storage schema.

The user would be required to describe input data and targetformats. Given these descriptions and a file matching theinput data description, the TRANSLATOR would producethe corresponding file in the target format. These descrip-tions would probably be registered with the image and couldbe changed by the user or data custodian by as simple amechanism as changing their registration parameters. Theseparameters might even be changed from one session toanother as the user invoked multiple native environments.

Using the primitive descriptions, aggregate data types couldbe described. For example, an N-element, floating-pointvector could be described as the concatenation of N floating-point representations. A data base record could be describedas a structure of primitive representations.

Queries might be returned as simple relations, or “flat files”with a standard format such as number of records, number offields, field names, and the row major order string of thequery content. The “data description language” representsyet another canonical or neutral language that protects usersfrom needing to know anything other than the characteristicsof their own native environment and the Accessor com-mands to communicate with other environments. It is likelythat current research on knowledge-based systems and auto-matic program generators will shed more light on how thismight best be accomplished. Synchronous access is a prob-lem of significant dimensions without an obvious answer.The problem arises whenever compound queries requireaccessing one data base for pointers to another, and inrelation to any form of update request. The confoundingissue is the need to keep several processes concurrent andsynchronized.

DBMSs devote a considerable amount of code to the syn-chronization issue in an attempt to avoid lockouts, to main-tain data integrity and to provide a consistent set of userviews. The same set of issues promoted from homogeneousto heterogeneous systems are all the more difficult. Thedifficult nature of the problem can be seen in so simply stateda problem as creating a relational join from two or more datasources managed by two or more DBMSs. I am not con-vinced that such problems are intractable, only extremelydifficult.


The solution may be found in some form of shared file systemwithin the image or intermediary (auxiliary) processors thatpermit processes to share the same variable (a data item forinstance). There are models of such systems, but the ones Iam aware of maintain homogeneity at least at the level of theoperating system.5

High Performance Computing

It is not difficult to imagine that there are many researcherswho could profitably use the power of a supercomputer butwho have never considered the option or have rejected itbelieving the threshold to use one is too high. One approachto lowering the threshold for such potential users wouldinvolve making access to the supercomputer appear a simpleextension of their native computing environment. Such anextension could be implemented via electronic mail com-bined with sufficient intelligent routines available at somesupercomputer site or hub. As many potential users might notbe FORTRAN programmers in their native computing envi-ronment; significant lowering of the access threshold wouldneed to include a problem description “language” which didnot involve mastery of FORTRAN or its nuances for anyparticular supercomputer system.

Such a strategy would require attention to current candidatenative computing environments as well as such environ-ments that might develop in the future. For instance, wecould imagine a spreadsheet user describing an extremelycomputationally-intensive mathematical programming orheuristic search problem in some spreadsheet format. Theconstruction of an “interpreter” to restate the problem im-plicit in the spreadsheet format can be theoretically envi-sioned, but seems hardly worth the effort given the amountof additional information the user would need to supply in(probably) some format other than the spreadsheet. Whenextended to the other (perhaps hundreds) candidate nativeenvironments the task appears formidable indeed.

An alternative strategy would be to develop some form ofcanonical language that requires far less time and effort tomaster than supercomputer FORTRANs, which can itselfreference computational data maintained in the user’s nativeenvironment. Efforts designed to produce similar types ofcanonical language environments are exemplified by thework on the Initial Graphics Exchange Specification (IGES)and the Manufacturing Automation Protocol (MAP). Thecanonical language envisioned here, while serving a similartype of exchange function, is not directed toward data, butrather toward metadata and process—that is, toward de-

scriptions of data and how those data are to be processed. Itis also dissimilar in that the exchange moves only in onedirection, from the user in his or her native computingenvironment toward the supercomputer that is the targetprocessing system.

A canonical language would have an additional benefit inthe avoidance of the user’s need to know in advance thetarget supercomputer. In this way the nuances of problemdecomposition to effectively use the form of parallelism orvectorization implemented in a particular system and com-piler would be avoided. The user would provide a “generic”description of the problem to be solved and the canonicallanguage would be translated to a “proper” set of FORTRANcode once the target machine was chosen. This would permita departmental hub to provide value-added service onseveral (perhaps radically dissimilar) supercomputers. Onecan imagine several scenarios of least cost, least wait, etc.path algorithms used in selecting the target supercomputer.

The forms such a canonical language might take are manyand varied. Several candidate approaches come immedi-ately to mind.

• A natural language interpreter

• Adaptation of a highly parallel processing language suchas APL

• A semi-graphical block flow language that describeddata transformations desired

• A FORTRAN-like pre-processor that invoked the namesof generic FORTRAN subroutines

The enterprising linguist could probably add several more.

5 James H. Morris et al, “Andrew: A Distributed PersonalComputing Environment,” Communications of the ACM,March 1986, pp. 184-201, is an innovative example underdevelopment at Carnegie-Mellon University.

he discussion in this section has necessarily been tenta-tive and not prescriptive. Implementation of the initial

phase of the image might well change the nature of many ofthe issues referred to here. I am convinced that a knowledge-based system will ultimately arise in response to the difficul-ties implicit in a heterogenous, distributed computing envi-ronment. Its exact form is far from obvious to me at thisjuncture. I have attempted to survey some of the longer-termissues only to give a flavor of where I see the single systemheading.

T


3A NEW PARADIGM

We have found a strange footprint on the shores of the unknown. We have devised profoundtheories, one after another, to account for its origin. At last we have succeeded in constructingthe creature that made the footprint. And lo, it is our own.

Arthur Eddington

As the paradigm for campus information systems turns froma mainframe-centered, Ptolemaic model to a user-centered,Copernican model, it will be necessary to redefine responsi-bilities.

As the strategy evolves, we should find the central computingoperation less and less involved in the user’s native environ-ment and increasingly concerned with image-associatedactivities. This suggests that it should be withdrawing from anumber of its current user service activities and moving intonew areas. For instance, classes and consulting activitiesrelated to native user environments should be decreasingwith emphasis being switched to development of the singlesystem image. For the near future, terminal-based officeautomation will probably continue to be a popular nativeenvironment for casual users, but it should be supported as

shift in staff expertise. Systems programming skills willbecome increasingly important, and user consulting skillsless needed. At the same time, systems programming skillswill be required for an even greater diversity of hardware andsoftware configurations.

A similar analysis of skills within software developmentcould be made. We can expect decreasing need for con-structing DBMS procedures and screen interfaces and in-creased emphasis on the conceptual model of data sup-ported within the institutional systems. Interfaces to theimage in the form of TRANSLATOR and ADMINISTRATORrules will become increasingly important. Software develop-ment organizations should be able to expend more effort onmetadata issues and the development of meta-operations ondata dispersed throughout the institutional data base.

Figure 3. Responsibility Matrix

Institutional

Departmental

Personal

College

MinicomputersCAD/CAMLaboratory AutomationFaculty Consul

PC NetworksPC Labs

Department

TerminalsFaculty PCsPC SoftwareStudent Instruction

Institution

Image E-Mail Archives PrintingComputer Network

Office AutomationSupercomputingLANsSpec. SoftwareLibrary Automation

Device ConnectionMaintenance ContractSite LicensesPC Demos

a departmental computing activ-ity. Institutional consulting relativeto personal computers should bemoving in the direction of assis-tance in hardware and softwareselection and configuration. Theinstitution should be phasing out ofany involvement in operating per-sonal computer laboratories as stu-dent-owned machines proliferate.This type of analysis could be ex-tended almost indefinitely, butconsideration of the following fig-ure will suggest the general natureof the shift of activities that shouldbe occurring at the institutionallevel and suggest how thesechanging responsibilities mightdevelop.

Accompanying such a shift in em-phasis should be a commensurate


In a similar vein, those organizations concerned with com-munications—voice, data, and video—will need to cometogether to develop a common planning framework. A set ofindividual foci on telephones, on data communications, andon video will not be found cost effective or sufficient. Thecurrent push for all digital networks, particularly ISDN anddigital television, will help focus an interest in seeing thenetwork physical medium as the same for all three technolo-gies. The result is likely to be increased use of fiber optics andmuch higher bandwidths to accommodate a burgeoning setof high speed departmental LANS.

Institutional operation of a digital PBX or acquisition ofsimilar services from the local common carrier will be seenas increasingly attractive. However, even with ISDN suchsystems will only address low bandwidth data traffic and willnot address video issues. The rationalization of high speed(megabit) LANs will be one of the major problems facinginstitutional communications organizations.

Libraries face the most difficult challenge presented by fullrealization of the information age. Libraries have over a2,000-year history that, with the exception of the inventionof printing, has changed little. Electronic publishing presentsa set of challenges that will be difficult for libraries toassimilate. The change agents are computers, communica-tions technology, and a plethora of new storage media. Theproblems for libraries will be exacerbated by continuinglegislative confusion over copyright law and the need forpublishers to decide how they intend to deal with theinformation age themselves.

However, it is the integration of the library’s stock of informa-tion that represents the fundamental step for the singlesystems image. Clearly the one addition to the campusnetwork required to make it truly ubiquitous is bibliographicand full text information from the campus library and na-tional data sources. In the longer term, the library might wellaccount for the vast bulk of traffic in the network and themajor load on the image.

In one sense the library represents only a specific instance ofanother data base participating in the image. In anothersense, it is a very special problem because of its size andrelative isolation (at least at present) from most campuscommunication networks. It is not difficult to envision that,once the major communication problems are addressed bythe image, the library will become the major focus ofcomputing and communications interest.

The library is likely to be relatively unique on the basis of datatypes as well as volume. As the library becomes a repositoryfor machine readable data, the text, graphics, and bit-mapforms of data are certain to become significant componentsof the data types flowing through the image. Transmission ofthese data types will require significantly higher bandwidthsthan electronic mail or even data base queries.

An additional confounding factor will be the high use oflibrary resources by off-campus and, frequently, non-univer-sity users. Many of these complications are obvious. One thatis sure to present conceptual difficulties is the question ofcost recovery. While it is entirely appropriate for the univer-sity to treat the image (and its associated computational andcommunication resources) as an operational overhead item,substantial usage by non-university community users willgenerate the need for cost recovery.

To this point, all questions of managerial accounting in theimage have been ignored. As the strategy is implemented, itwill be necessary to make a number of adjustments in bothfunding (amounts and distribution) and chargeback policies.A user acquiring a personal computer can expect to payabout $3,000 for his or her native computing environment—charges for access to the communication network (primarilysome form of modem) and CPU resources are added to this.A user acquiring a terminal can expect to pay about $700, notincluding communication connectivity and CPU resources.Current pricing policies at most institutions favor usersacquiring terminals in a rather significant way. Not only is theout-of-pocket expense several thousand dollars less, butCPU resources are generally available in effectively unlim-ited amounts at no cost to the user. Such a policy favors useractions that are exactly the opposite of what is generallydesired.

It seems an appropriate time to reconsider the chargebackpolicy for computing resources. Dividing the issue into twocategories, subsidized and cost-recovered, suggests thatthose services shown in the upper left box of Figure 3 shouldbe subsidized, all others cost-recovered.

SUBSIDIZED—resources and services associated with thesingle system image and information center-like activi-ties.

RECOVERED—resources and services provided (or facilitiesmanagement costs) to support mainframe- or mini-based native environments.

For instance, users of mainframe-based systems should beexpected to pay for services received. On the other hand,personal computer or College-owned minicomputer usersshould not have to pay for services provided by the image,although when the service results in some “hard” form ofoutput, it might be reasonable to charge for the “pieces”produced. As a general principle, access to the communica-tions network (beyond the local modem) is subsidized;general purpose computing resources are cost recovered. Aswith any rule, there will likely be compelling reasons forexceptions, but they should be clearly recognized as such.Putting principle into action, say for office automation users,this would mean that the cost to provide the service shouldbe determined and the cost recovered from users on usage


basis. If office automation services can be produced morecheaply (or the service level is significantly higher) viamainframe, then users desiring an office automation-likenative environment will opt for mainframe-based service. Ifnot, less expensive (or higher service level) approaches onpersonal computers will be chosen. A more equitablechargeback policy will allow computing services to be morenearly market driven, rather than subsidy directed as ispresently the case.

With such a view, it would appear that current subsidy vs.recovery policies for most institutions are nearly reversedfrom where they should be. Development of the communi-cation portion of the infrastructure should, as a generalprinciple, be subsidized. Institutionally maintained nativecomputing environments, as a general rule, should be costrecovered. One way of seeing the issue is to recognize thatthere is frequently more computing power purchased byindividuals or acquired by grant, considerably less powerprovided as a subsidized commodity by the institution. Thisimbalance can be expected to grow significantly, evenwithout changes in cost recovery policies.

An immediate tactical question should be a study of currentchargeback policies with the intent of defining a policy more

consistent with desired user behavior and real costs. Aprincipal issue in such a policy study should be the distribu-tion of chargebacks between subsidized and “real” dollarsfor the various services of the computing center, communi-cations network, and systems development activities.

An ancillary question to be answered would involve thedistribution of “real” dollars—between the computing cen-ter, communications network, systems development, andcolleges and departments. While the historical record atsome universities generates some misgivings regarding theplacing of “computing” dollars in unit budgets with theability to use them for travel, paper clips, etc., there ispotentially some merit in allowing “computing” dollars to bespent for personal and/or regional computing services in lieuof global services. Such an arrangement would have to bewell planned and slowly phased into, but might serve tomuch more accurately register user demand.

Our goal should be the creation of a single system imagethat provides the ubiquity and simplicity that we

associate with the telephone system. Such an effort willrequire shifting many current burdens from the user to aknowledge-based network and rationalizing the way theuser relates to the services provided.


CAUSE is a nonprofit professional association whose mission is to promote effective planning, manage-ment, development, and evaluation of computing and information technologies in colleges and univer-sities, and to help individual member representatives develop as professionals in the field of informationtechnology management in higher education. Incorporated in 1971, the association serves its membershipof more than 740 campuses and 2,000 individuals from the CAUSE national headquarters at 737 Twenty-Ninth Street, Boulder, Colorado 80303. For further information phone (303) 449-4430 or send electronicmail to [email protected].

C USEA

Documents

A Single System Image: An Information Systems Strategy