Project Athena:

Supporting distributed
computing at MIT
by J. M. Arfman
P. Roden

Project AthenarMwas an educational computing observed that there was very little, if any, com-
initiative at the Massachusetts Instituteof puter usage for educational purposes on the un-
Technology, undertakenin partnership with the
IBM Corporation and Digital Equipment dergraduate level, compared to that in graduate
Corporation from 1983to 1991. This paper gives education and research. The study originated in
an overview of the network-based distributed theSchool of Engineering, and DEC wasap-
computing services, developedfor a number of proached asthe single partner.However,the
UNIP-capable workstations. These servicesare recommendation was soon made to createan ex-
extensions to the native operating systemsof the
workstations, and provide interoperability as well tended project that would provide an undergrad-
as systems administration facilitiesin a large uate computing environment to benefit all of the
heterogeneous workstation environment. Under MIT undergraduate curriculum. At the same time,
Project Athena, a mature distributed computing additional sponsorsweresought,and IBM and
environment was developed. Its organization and
support structure maybe used as a model when DEC officially became partners. Donations from
planning a new installation, whether on a many other companies andindividuals were also
university or commercial campus. A sectionof received, particularly in the early years.
this paper deals with the support requirements
for distributed computing environments, based The designersof the Project Athena system chose
on the Project Athena experience.
to implement a uNIX**-based distributed work-
stationenvironment.Thisdecision was made
knowing full well that this technology was not
available in the marketplace at the start of the
project. The goals were to create a computing

P roject Athens**, the Massachusetts Institute

of Technology’s widely known model of dis-
tributed computing, was launched in 1983 by MIT
environment that would scale up to 10 000 work-
stations and accommodate heterogeneous hard-
ware, but yet be “coherent.” This concept means
in partnership with the IBM Corporation and Dig- that a user could go to anyworkstation and access
ital Equipment Corporation (DEC) to address the any files or applications withoutfinding major dif-
Institute’s need to provide widely available com- ferences in the user interface and service deliv-
puting resources to the campus. The project part- ery. The Berkeley UNIX base was chosen orig-
nership ended on June30, 1991, eight years after
it began. The Athena** computing environment Wopyright 1992 by International Business Machines Corpo-
ration. Copying in printed form for private use is permitted
has been officially adopted by MIT as its infra- without payment of royalty provided that (1) each reproduc-
structure for delivery of educational as well as tion is done without alteration and (2) the Journal reference
selectedresearchandadministrative applica- and IBM copyright notice are included on the first page. The
tions. title and abstract, butno other portions, of this paper may be
copied or distributed royalty free without furtherpermission
by computer-based and other information-service systems.
An MIT study in the early 1980s on the use of Permission to republish any other portion of this paper must
computer technologyin undergraduate education be obtained from the Editor.

550 ARFMAN AND RODEN IBM SYSTEMS JOURNAL, VOL 31, NO 3, 1992
inally to achieve this coherence on the IBM and dustry standard. More recently, the Kerberos**
DEC platforms, with IBMi4.3 on the RISC Technol- authentication mechanism, first implemented at
ogy (RT*) and 4.3 BSD** (Berkeley software dis- Project Athena, has been adopted (with Hewlett-
tribution) forDigital’s workstations. The current Packard extensions) by the Open Software Foun-
and future environment will include Advanced dation (OSF) for inclusion in its Distributed
Interactive Executive (AIX*) for IBM’s (Personal Computing Environment (DCE**). Threeother
System/2* (PS/~*)and RISC System/6000*, Ultrix** technologies were submittedby Project Athena to
for DEC’S workstations, A/ux**(AppleUNIX) for the OSF in response to its call for Distributed Man-
Macintosh**, and others. agement Environment (DME* *) technology:
Moira**(service configuration management),
The concept of the public workstation was also im- Zephyr** (on-line notification), and Palladium**
portant as it freed the users and their files from be- (print services), the latter together
with IBM, DEC,
ing associated with unique physical workstations. and Hewlett Packard.
This system would be secure, “location-indepen-
dent,” centrally administered but physically dis- The Athena computing environment
parate, and provide a consistent graphical user
interface. It would be easy to install and update The following sections are devoted todescribing
software. The concept had far-reaching implica- the physical layout and architectureof Athena, as
tions for system design. It took almost five years well as the services provided.
to make it a reality, and was the reason for the
extension of the original project from five to eight The MIT campus network. Project Athena does
years. The last three years of Project Athena were not manage MITnet (MIT network), the networkon
dedicated to improving and solidifying the system which it relies to distribute traffic. However, the
infrastructure. As a result, computer usage at MIT impetus for MITnet did come at least in part from
has increased beyond all expectaiions, and the therealitythatthesuccess of ProjectAthena
impact of this technologyis growing and receiving would depend on reliable transmission capabili-
worldwide recognition. ties over the entire campus,as well as beyond it.

The system is totally transparent to the user.

Proj- MITnet consists of a 100 Mb/sec fiber optic FDDI
ect Athena’s services make it possible to go to (fiber distributed data interface) spine, extending
any workstation and access files and applications the entire length of the campus, or almost two
that can be resident anywherein the system. The miles. The spine does not have any systems di-
user provides a logon userid and password; the rectly attached to it, as shown abstractlyin Figure
system servicesrecognize and authorize the user 1. Rather, it has a number of routers to which
to use the system. various subnets are attached.

MIT’S students have embraced Project Athena en- MITnet is connected to therest of the world via the
thusiastically. Over 95 percent of all undergrad- New England Academic and Research Network
uates are regular users. Athena has had a signif- (NEARNET), which is one of the midlevels con-
icant impact on how they work and on how they nected to the National Science Foundation Net-
interact withfellow students and faculty. A small work (NSFNET). MIT has been awardedacon-
but growing number of faculty feel that Project tract to bethe Nodal Switching System (NSS) for
Athenaalso changed their teaching methods, NEARNET, and is directly connectedtothe
making their face-to-face contact with the stu- NSFNET backbone. A cluster of Athena worksta-
dents more productive, and helping them develop tions was installed 80 miles away from the MIT
intuitive learning in their students. campus at the Woods Hole Oceanographic Insti-
tute, and is connected to MITnet by a series of
In retrospect, the challenges wentbeyondthe microwave links. MIT’s connectivity creates
scope that was originally envisioned. On the other “communities of scholars” electronically. Stu-
hand, the significant accomplishments at the sys- dents and faculty have embraced this method of
tem design level were also far beyond anything sharing information on a global level. The approx-
expected at the time. For instance, X Win- imately 4000 systems at MIT, ranging from per-
dows**, which camefrom MIT, led to the estab- sonal computers to a Cray-2* *, are connected via
lishment of the X Consortium, and is now an in- Ethernet subnets, attached to the spine via rout-

IBM SYSTEMS JOURNAL, VOL 31, NO 3, 1992 ARFMAN AND RODEN 551
 c
4POSTOFFICE 1
[ OLC l-
(OLTA +
PRINTSERVER
( r F
+G")
-(-)

CENTRAL SERVICES
SUBNET

552
 SUBNET

USER FILES
COURSE
LOCKERS

RESEARCH
MACHINES
t J

\
CRAY 2

IBM 3090
SUBNET

Redrawn with permissionof the Massachusetts Instituteof Technology

ware has to be provided rapidly on demand. Each now serves the whole MIT campus as a delivery
subnet has one of these servers, with the binaries system. In 1989, a decision was made to grant
for every type of workstation present on that par- graduate students Athena accounts. As of No-

I
ticular subnet.
vember 1990, about 70 percent of all graduate
stu-
dents had accounts, in addition to 98 percent of all
While the initial focus of ProjectAthena wason undergraduates. The IBM andCraysystems,
undergraduateeducation,the implementation which are used foradministrativeandresearch

IBM SYSTEMS
JOURNAL, RODEN 31, NO 3, 1992
ANDVOLARFMAN 553
 purposes, are also accessiblefrom Athena work- User files, courseware, and licensed software are
stations. Someone who has an accounton one of stored on the file servers in “lockers,” hierarchi-
these systems can access mainframe applications cal directory structures, e.g. NFS mountpoint or
within an Athena window.MIT’S strategic goal for AFS volume. Eventually, both the RVD and NFS
file service for users’ home directories and system
software delivery will be replaced by AFS. The
user file and course locker servers are shown at
the upper right in Figure 1. The total amount of
backed-up storage on these servers exceeds 40
The “dataless” workstation gigabytes. At present, students are given 1.2 meg-
model allows a user abytes of backed-up file space, with more avail-
to log on anywhere able if the user canget approval or iswilling to do
a private backup. Thisis not so much alimitation
on the network. in the systemdesign as a financial decision based
on the number of users and the number of file
servers available, taking into account the avail-
able backup facilities, technology, and personnel.
distributed computing is to make the Athena com- Centralservices. Aclient-server application is
puting environment the service delivery mecha- considered a “service” offered on the network.
nism for all instructional, research, and adminis- Someservicesarerequiredforthesystemto
trative applications, such as the library catalog, function, such as authentication, and others are
the Registrar’s Student Information System, and value-addedservices,such asthe on-line help
others. system.Theseservicesarecentrally adminis-
tered, butare physically located so as tominimize
File servers. The user and courseware file servers, impact on users, in the event that portionsof the
as well as those for central services, can be lo- network are unavailable.
cated anywhere on the network. Project Athena
initially adopted, with security modifications, the The Central Services, shown in the left half of
NetworkFileSystem (NFS**) fromSun Mi- Figure 1, consist of the suite of Athena network
crosystems, Inc., an industry standard. Athena services, developed over the lifetime of Project
also uses the Andrew File System ( U S * * ) from Athena. Together, they constitute a complete dis-
Transarc Corporation, and a migration to AFS is tributed systems environment, which is scalable
under way. Thisfile system, originally developed up to at least 10 000 workstations. The architec-
by Project Andrew at Carnegie Mellon Univer- ture of the Athena system is shown in Figure 2,
sity, has superior distributed file management, layered on top of the operating system and com-
performance,andbackupcharacteristics. The munication protocol. While the systems services
thirdtype of server, the Remote Virtual Disk are shown together in Figure 1 for the sake of
(RVD), was developed atMIT in the Laboratory for schematic clarity, this does imply
not that there is
Computer Science to provide a fast, read-only, only one server of each type, or that they areall
binary server for each subnet. Typically, the sub- located on thesame subnet. On the contrary,it is
net RVD contains the code for the operating sys- necessary for continuous availability to have at
tems of the workstationsin that particular subnet. least the crucial services replicated acrossthe
network. Other services require multiple servers
Since Project Athena has adopted the “dataless” because the number of servers grows with the
workstation model, and a user can log on any- number of users.
where on the network, the location of the user’s
files is irrelevant to theindividual and is managed Kerberos. A consequence of the public worksta-
through a mapping from logical to physical ad- tion model is that authentication, or identification
dresses by the Hesiod** name server. attach The of the user, is no longer relevant at the worksta-
command is an abstraction thatallows the user to tion level. All public workstations are considered
be unconcerned about the actual physical loca- “untrusted,” to the extent that the root password
tion and structure of the files. is published in all documentation. To authenticate

554 ARFMAN AND RODEN IBM SYSTEMS JOURNAL, VOL 31, NO 3, 1992
Fiaure 2 Project Athena system architecture

i
APPLICATIONS

I ON-LINE SERVICES
(DOCUMENTATION, CONSULTING)
I
I ELECTRONIC MAIL
I
(DISCUSS, NETNEWS) (LPQ/LPD, PALLADIUM)

I
II
I
WINDOWING SYSTEMS (X, MOTIF, ETC.)
I I

/
DISTRIBUTED FILE SYSTEMS (RVD, NFS, AFS)
I I

(HESIOD) (ZEPHYR)

TCP/IP NETWORK TRANSPORT

. OPERATING SYSTEM

NETWORK

I I

Redrawn with permission of the Massachusetts Instituteof Technology

the user, rather than the system, the Kerberos Hesiod. In order to associate the user with his or
authentication system, basedon the trusted third- her own tailored environment and home direc-
been
party
has designed and imple- tory, a mapping function is required that trans-
mented at Project Athena. The user’s password is lates logical into physical addresses. Since the
encrypted, so that no password will enter the net- user can be at any workstation, only the authen-
work in the clear. The secure Kerberos server ticated userid for that session is of importance.
thengrantsencryptedtickets, valid for a few The Hesiod4 server updates its information from
hours, which identify the user as legitimate to the Moira database(describedlater),andper-
Athena network services such as file servers. All forms the required mapping when the user signs
of this is transparent to the user beginning with on. As a consequence, the user is always assured
the logon process. of a completely personalized workstation envi-

IBM SYSTEMS JOURNAL, VOL 31, NO 3, 1992 ARFMAN AND RODEN 555
 ronment, with home directoryand other relevant Project Athena. Once a distributed environment
files attached as if they are stored locally. reaches more than 200 installed workstations, it is
essential to provide an economicalway to install
Post Office. In addition to Kerberos and Hesiod, andupdate client andserversoftware. Moira
a numberof other central servicesare available to makes useof a centralrelational database, where
the users via the network. The Post Office’ serv- all system control and configuration information
ers provide the store-and-forward capability re- is kept. It contains the authoritative data about
quired for mail, since the user no longer has a files, users,services,hosts, mailing lists, and
permanent physical workstation address. At any group lists. Changes in the database are propa-
time during a session, the mail that has been kept gated as service-specific configuration files to ap-
stored in the Post Office server may be requested propriate servers on a regular basis. Project Ath-
and transferred into the user’s home directory. ena’s model is influencing thedirectionand
standards for the industry.
Zephyr. In addition to mail, there is an instanta-
neous notification service called Zephyr,‘j which
allows users or services to sendmessage
a to one Dialup. A number of workstations areconfigured
or more individuals who arelogged inat the time. as Athena time sharing systems. Any terminal
A small Zephyr window, or Zephyrgram, pops emulation program, suchas PC communication or
up, containing the message. It is also possible to network remote software for logging in, may be
subscribe to “instances” or classes of Zephyr used to access this service. Analog connections
messages, so that a user who has subscribed are established by dialing into the modempool of
to all
the MIT digital telephone switch. These are then
messages on a given subject can receive these
routed through the telephone system to an X.25 PAD
automatically. Those not subscribed will not re-
ceive the messages. (Packet Assembly/Disassembly) andonto one of
MITnet’s Ethernets. Any available dialup server is
assigned, unless a specific type is requested. Au-
On-Line Consulting. On-Line Consulting’ (OLC),
which is built on top of the Kerberos, Hesiod, thentication is provided at the host workstation
level. Only those applications and services that
Zephyr, and Post Office services, allows the user
to send an instantmessage to a group of consult- do not require X can be accessed. Dialup is pop-
ants, to ask for assistance or information about ular for remote accessto mail and other common
anything regarding usage. Morethan 3000 of services.
these questions are answered every month, of
which 95 percent are answeredin under five min- Othercommonservices. A discussion of other
utes. These questions are alsologged and edited common services follows.
into the “stock answers” database. If the user is
no longer logged in when the answer becomes
available, the system reverts to regular electronic Discuss. Electronic conferencing, or Discuss**,
mail for delivery. Consultants can also be con- allows for both public and private meetings. It
provides for authentication, moderators,mail ar-
tacted via telephone or in person, but experience
has shown that electronic assistance is far more chiving, and transaction flags. There is no equiv-
alent supported product available today on het-
popular.
erogeneous hardware platforms.
On-Line Teaching Assistant. A variant of OLC is
On-Line Teaching Assistant(OLTA). This service Printing. Distributed andauthenticated print
was inaugurated in September 1990, to provide spooling, and location-independent printer con-
students with direct access to the teaching assis- figurations are provided. In addition, user quotas
tants for the courses in which they are enrolled. for printing havebeeninstituted.Thecurrent
It has been animmediate success, with overhalf quota is 1200 free pages per year. Students must
the freshmenusing it. About tenof the large fresh- pay for any excess printing through the regular
man courses utilized OLTA during the 1991 aca- student bill. This prevents the use of printers as
demic year. free copying machines. The Palladium print sys-
tem, jointly developed with IBM and DEC, was
Moira. Moira’s service management system’ pro- accepted by OSF for DME, and may eventually
vides the central management facility for all of replace Athena’s current print system.

556 ARFMAN AND RODEN IBM SYSTEMS JOURNAL, VOL 31, NO 3, 1992
On-line documentation.Information about all as- courses at MIT are now using applications devel-
pects of the systems is carefully maintained and oped with Athena MUSE. I1-l3
available on line. Searches can be doneby either
menus or keywords. The data are stored in a hi- The strengthof Athena MUSE is that the language
erarchy of small, logical modules. There is a sin- is extensible, i.e., the higher levels are writtenin
gle documentation source pool for both on-line terms of the lower levels of language. Also, it is
and hardcopy viewing. possible, depending on the author's interests and
level of expertise, to enter Athena MUSE at the
Map-based applications. Campus resources, lower levels. To the greatest extent possible, Ath-
such as available workstations in public clusters, ena MUSE has been made independent of the mul-
or offices, buildings, and libraries, can be located timedia platforms onwhich it resides, making it a
on electronic maps. Theapplication Xcluster dis- relatively simple matter to port it to new UNIX
plays a map of the campus and the location of operating systems and hardware.
each public cluster. When a particular location is
clicked on with the mouse, the number and type Athena MUSE is not dependent on the Athena dis-
of workstations currently free, well as as thenum- tributed environment, although it has been im-
ber in use, isdisplayed. The statusof each printer plemented to support Athena at MIT. The hard-
in that cluster is also shown. This is very useful ware included IBM RTs and DEC workstations,
because as students move from class to class, with the Parallax** video digitization board, and
they can plan where they could next use a free IBM Megapel displays. Video is supplied by the
workstation (of a specific hardware type if de- campus video cable or video disks. In 1991, work
sired) to continue their work. was completed by Bob McKie l3 at theCambridge
Scientific Center to port the AthenaMUSE capa-
Global messaging. This provides a mechanism bility to the IBM PS/2 with AIX and "Motion.
for disseminating messages to various classesof
users. For instance, a message could go to all The work done by the VCG is the basis for the
users in the freshman class to alert them to new Athena MUSE Consortium established at MIT as
course offerings. part of the new Center for Educational Comput-
ing Initiatives, under the direction of Professor
Steve Lerman. TheMIT members of the VCG form
Multimedia projects the core of the Athena MUSE Consortium staff.
The Visual Computing Group (VCG) at Project
Athena occupied a very special place within the Usage
project for the lasthalf of its existence. Members
of the group werelooking ahead to thetime when As of June 1991, there were about 11000 users
capabilities of the hardware and advancesin tele- registered, with over 3500 individuals accessing
communications would allow distribution to the the system at least once a day.Overall usage has
individual workstation of multimedia objects and risen dramatically over the last two years, with
applications. Theyenvisioned the need for a mul- each semester's statistics surpassing thoseof the
timedia authoringenvironmentthat would be previous one. An average of 100 regularly sched-
easy touse and flexible enough to accommodate uled courses each semester now require the use of
awidevariety of applicationsandauthoring Project Athena. About two-thirds of the 1300 or
styles. The prototype language or (maybe more so workstations are available in public and de-
accurately) environment they developedis called partmental clusters, the rest are deployed as pri-
Athena MUSE**. l o Thisprototypewastested vate workstations. As an early experiment, some
against a great variety of real applications. It en- of them were placed in student living quarters;
compassed education and other applications. As othersareused in electroniclecture halls and
a result, Athena MUSE has been enriched to the classrooms.
point where it is necessary tore-architect andadd
functionality, particularly in the area of multime- While studies have been made of student usage
dia editors. There has been worldwide interest in patterns, there have been none in the past two
this work, particularly because it has been dem- years, during which the utilization time has in-
onstrated to thousands of visitors, and several creased rapidly. Past questionnaires have shown

IBM SYSTEMS JOURNAL, VOL 31, NO 3, 1992 ARFMAN AND RODEN 557
 that word processing and electronic mail tended Steve Lerman, thereis a focus fornew academic
to predominate initially, accompanied by simula- computing initiatives at MIT, which includes in-
tion and games. The actual use of course mate- dustrial partners in many of its projects.
rials was not as heavy as had been hoped. This
can be accountedfor partly by realizing the mag- Much of the distributed computing technology
nitude and difficulty of writing and maintaining developed at Project Athena has been adopted or
courseware for a noncommercial operating sys- submitted to OSF and will become available in
tem, since third-party tools are available.
not Sec- vendor products over the next few years. Uni-
ond, more funding for courseware development versitiesand commercial customers are begin-
was available during the early years of Project ning to look for this technology to solve their
Athena, when the operating system wasstill un- growing needs for system management and in-
der development. Much of the courseware was teroperability in heterogeneous
a distributed
written by students, and therefore theoriginal de- computing environment.
velopers were often nolonger available at MIT to
make the necessary changes when a new release
of the system software required them. Third, im- Support requirements for DCE
portation andporting of suitable softwarewas not The objective of this section is to provide DCE
always possible or affordable. planners with realistic support estimates, based
on the customer’sown general requirements and
There has been recent, qualitative evidence that environment.
the usage is starting to shift somewhat from writ-
ing papers,sending mail, and playing games Organization. Using the current Project Athena
toward scientific modeling, visualization,and organization as a model, reasonable assumptions
courseware use by students, indicating a higher can be made concerning which parts would be
level of maturity in the integration of Project Ath- necessary in order to support the basic installa-
ena into the undergraduate curriculum. Another tion. These assumptions were arrived at with the
observation is that many students like working in assistance of the MIT staff. Project Athena went
theworkstationclusters,wheretheycanask through extensiveself-examination in order to de-
questions, work together, and learn from each termine what parts were to be merged with its
other. Information Systems organization after the for-
mal end of Project Athena in June of 1991. Four
Athena resources are made available to the fac- functional areas are required to support the in-
ulty to help keep courseware current. Themigra- frastructure:requirementsand initiatives, user
tion to vendor-supplied operating systems, the services, systems and operations, and adminis-
expanding suite of third-party software, and the tration. The titles of the functional areas are ge-
stability of the authoring environment, will addi- neric, and do not necessarily represent actual de-
tionally have a positive influence on courseware partments.
import, creation, and usage.
Requirements and initiatives. This area consists
of faculty or management liaison, third-party soft-
The future wareand application support,authoring tools,
MIT made the crucial decisionin the fall of 1990to and productivity tools.
make the Project Athena infrastructure part of
MIT’S basic service delivery organization, thereby 9 Facultyor management liaison-Users and
insuring that thelegacy of Project Athenalives on their management are provided with informa-
at MIT. This decision was made as a result of a tion vianewsletters, meetings, etcetera. The
very comprehensive and detailed year-long study people in thisgroupare thecatalystsfor
by the Committee on Academic Computing in the change-whether acting as ombudsmenfor
O OS, composed of a broad selection of MIT fac- user representatives, or presenting new ideas
ulty, whichassessedtheInstitute’sacademic and facilities to theusers. They teach and assist
computing needsfor the next decade. Inaddition the faculty, and give advance warning of sig-
to educational computing, access tolibraries and nificant changes and improvements.When nec-
administrative services is now provided. Under essary, software development support and re-

558 ARFMAN AND RODEN IBM SYSTEMSJOURNAL, VOL 31,NO 3, 1992

 sourcesare
providedto
ensure
thatuser other hand, control over the accounts may be
applications are portable after system software much more important and require more timely
upgrades or changes. actionbecause of securityconsiderations.It
Third-party software and application support- may well be that the resources required in both
Commonly used
applications
areselected, environments turn out to be about the same
tested, and ported if necessary. These applica- when calculated proportionally.
tions may need to be replaced or upgraded from
time to time. This group makes sure that the Consulting-The degree to which consulting is
right applications are available to the users, us- available to the users depends on management
ing known user requirements as criteria for se- philosophy. In principle, all the tools areavail-
lection.Examplesare CAD/CAM applications, able, from personal contact, to assistance by
technicaltrainingpackages,foreignlanguage telephone,toon-lineconsultingandon-line
instruction, GIS (Geographical InformationSys- help. The latter does not requirea person to be
tem)software,ordatabasemanagementsys- at the other end, and is generally a menu-driven
tems. search program for information and answers to
Authoring tools-In most environments, users the most frequently asked questions. The hu-
want to or need to write their own software. It man consultants are trained in a wide varietyof
is the responsibilityof the requirements andin- skills,
with some necessary specialization.
itiatives group to support these efforts by se- They can range fromfull-time paid profession-
lecting,testing,andinstallingauthoringtools als to part-time volunteers. These may also be
that will be easy to use and that will result in distributed directly into the software support
quality software packages. organizations of vendor suppliers.
Productivity tools-These are a special set of
applications,whose goalis toincreasethe Training-User training can be implemented in
user’s personal and professional productivity. many different ways. In universities, it is most
Examplesarefound in office systems,word efficient to schedule the initial training when
processing, and spreadsheet applications. students arrive, and it pays to have a live in-
structor give out information to large groups.
User services. Thisareaconsists of userac- For faculty, training is often done on an indi-
counts, consulting, training, and documentation. vidual basis. While new usersin the commercial
environment need training on demand, a case
User accounts-The management of the user could still be made for periodic group instruc-
accounts may differ depending on the custom- tion, as well as self-study or on-line education.
er’s environment and policy. For instance, in
the university, about one quarter of the users Documentation-Documentation, both on-line
leave and about the same number enter during and hardcopy, is one of the top user require-
the academic year. Adding a significant number ments. This function requires a staff of people
of users can be done quite mechanically. For who understand the technical environment as
instance, MIT manages this by extractinginfor- well as the user requirements for lucid, well-
mation from tapes provided by the registrar and organized documentation. Although much of it
the payroll office. All the user has to do is to can be madeavailable on line, some basic man-
access the system by entering an ID, establish a uals have to available
be in hardcopy, to initially
password, and wait about a day for the network train the user, or to provide information in case
resources and configuration to be set up. De- of system malfunction.
leting users turns out to be much of more
a chal-
lenge, since many may come back as graduate Systems and operations. Thisareaconsists of
students. In any case, private files would be customer service, system support, release engi-
destroyed prematurely if users were removed neering, and software engineering.
as mechanically as they can be added.
Customer service-A staff has to be available to
In a commercial environment the user group plan, install, and repair the systems. They are
may be more stable than at universities, and not only reactive, but proactive, and constantly
adds, deletes, or changes usually do not happen monitor the overall andindividual status of the
at a special time of year or all at once. On the installedequipment.This is done in several

IBM SYSTEMS JOURNAL, VOL 31, NO 3, 1992 ARFMAN AND RODEN 559
 ways. The staff is capable of monitoring from a Assumptions. Certainassumptions had tobe
centralpoint.Inaddition,“clusterpatrols” made in order to make the rules we developed
visit the workstations about once a day, to solve generally usable.
obvious problems where possible, and report
those that require further attention. They are Telecommunicationsandnetworksupporthas
the interface to vendor service organizations, if not beenincluded. There are many university and
repairs or replacementscannot be done in- commercial customers who already have a tele-
house. communications network and associated support
System support-The systemsupport staff structure in place. It is therefore assumed that the
takes care of systems administration and en- network infrastructure lies outside the scope of
sures availability and performance of central these estimates.
network servers. File server capacity, systems
and user file backup are monitored.
Release engineering-This function is required The proposed DCE installation is on one campus.
even if only one operating system is installed. The assumptionis that all workstations arewithin
The Athena services have to be reintegrated for a reasonably small geographical area and are all
each new release or version of the operating served from the same high-speed network. For
system. When distributing software updates to instance, it should be possible to walk or bicycle
a large group of users, the potential for disaster to a workstation location within a few minutes,
is always present. Not just one user, but all not hours. An example of this is a single univer-
users will be affected by any error or code de- sity campus, or a group of commercial buildings
fect introduced by a new software release. Soft- which are within the same city andwhich have the
ware updates must be donein a controlled and same access to the company’s high-speed net-
tested way.When more than one operating sys- work.
tem is involved, release engineering becomes
one of the most essential systems functions. There is no application development. Commer-
The staff does the systemsintegration and test- cially available applications satisfy the need of the
ing, andensuresthateverythingremains in- customer. Of course thisis not a realistic assump-
teroperable. tion for most customers, but it would be impos-
Software engineering-New versions and re- sible to estimate the resources to support every
leases of software are a fact of life. As more and customer’s application requirements.
more of the operating systems are providedby
vendors, the taskof integrating the Athena soft- The space for workstations and servers has been
ware and building new releases is a constant allocated and all facilities arrangementsare in
one. To some extent, each Athena-like instal- place. It may be difficult to find and build suitable
lation will have to face this, since local modi- space for the DCE clients and servers. The esti-
fications will exist, and awide variety of work- mates in this paper do not take into account the
stations and associated operating systems may very considerable initial work requiredto find and
be installed. In addition, it is important that new fund the space required, both for the workstations
releases of the Athena suite of software be in- and servers,and the people who aregoing to sup-
corporated as they become available, in order port them.The assumption hereis that, once hav-
to take advantage of fixes and functional en- ing made the decision to install the DCE, these
hancements. resources are the same whether or not the Project
Athena model is chosen. This assumption may
have to becarefully examined by the customer for
Administration. Othercriticalbutoftenover- accuracy.
looked aspects of DCE support requirements are
the areasof budgets, inventories, and license and How to estimate support requirements. At first, it
contract negotiations, as well as other adminis- might seem simplistic to design a “formula” to
trative issues.Management and personnelfor this estimatesupportrequirements. In talking with
functional area have to be planned for, and re- the experts on the subject, managers, and per-
sources allocated. The responsibilities include fi- sonnel at Project Athena, it became evident that
nance,propertymanagement,contracts,soft- although everyone had an intuitive understanding
ware licensing, and administrative support. about the human resources required, it was dif-

560 ARFMAN AND RODEN IBM SYSTEMS JOURNAL, VOL 31. NO 3, 1992
ficult to translate this into real numbers when a newed, etc. The number of licenses may reach
new installation was discussed. Resources are de- several hundred. The number of contracts to be
fined interms of full-time employees. Thefollow- negotiated, and contract and
license renewals can
ing variables are relevant. be significant.

Workstations (clients only). The number of work- Resource estimate: W2.5, where L is the number
stations affects the resources required for tracking of licenses.
and inventory, hardware maintenance, installa-
tionhe-bootinghpdating, and system-wide serv- Vendor operating systems and applications
ices. (VOSAP). VOSAP meansthevendoroperating
system supplied for aspecific platform. New ver-
Resource estimate: W/500, where W is the num- sions and releasesare a factof life. One school of
ber of workstation clients. thought is that there is a nonlinear relationship
between thenumber of operating systemsand the
Users. The number of users affects accounts ad- resource required to support them, with a “wall”
ministration, usertraining, “how touse” consult- at about four or five. The activities associated
ing, documentation, and serverconfiguration and with VOSAP support are release engineering, cli-
sizing. entconfiguration, testing, documentation, and
defect tracking and reporting.
Resource estimate: U/lOOO, where U is the num-
ber of users. Resource estimate: V, where V is the number of
distinct vendor operating systems andtheir plat-
Clusters. Acluster is a physically co-located forms.
workgroup, with systems in open areas and/or
private spaces. Clusters are characterizedby dis- Combined requirements. When combined, the to-
tance between clusters, size (i.e., number of cli- tal human resources (HR) required are expressed
ents per cluster), and printer and system server in Equation 1:
requirements. These factors affect resources re-
quired for planning, deployment, paper and other HR = + U/lOOO + C/15
W/500
consumables,space management andsecurity,
and server administration. + At50 + W25 + V (1)

Resource estimate: C/15, where C is the number One additional piece of information is needed in
of clusters. order to use this formula for estimating resource
requirements. There are five sets of human re-
Supported applications. An application is defined source skills required to support the total envi-
as a licensed product or its equivalent if it was ronment. These are roughly related to each term
developed in-house. Examplesare:word pro- in the formula, with Aand L combined. Themin-
cessing, a CAD application, or a courseware pack- imum value of HR is therefore five. Depending on
age. Thenumber of applications provided and the environment and how resources are counted,
supported centrally, exclusive of development, customers are reporting anywhere from 10 to 25
affects the resources required for compilation (if workstations supported per support resource in a
porting is required), building the software distri- non-Athena installation. This brings us to the con-
bution, integration and testing, installation, up- clusion that a Project Athena-like environment
dates, software maintenance,application-specific only starts to pay offin terms of resources re-
user support, and higher-level documentation. quired to support it when the number of client
workstations is somewhere around 150.
Resource estimate: N 5 0 , where A is the number
of supported applications. An important factor to keepin mind is that these
estimates are based on support requirements for
Licenses required. A licenseis defined as the right a mature installation. It is quite possible, when
tousethe application softwarefor potentially starting anew installation, to get by with far fewer
multiple users onmultiple platforms. In all cases, support people. In this case the favorable factors
atleastonecontracthasto be negotiated,re- are that all hardware was provided by a single

IBM SYSTEMS JOURNAL, VOL 31, NO 3, 1992 ARFMAN AND RODEN 561
 ~~ ~~~ ~ ~

Table 1 Test of the support formula parameters, showing actual and calculated resources

Profiles of Six Athena Installations

Parameter Divisor MIT NCSU ucsc UMD ISU UMICH

Workstations
1300 500 350 200 20 850 265
Users 11000 1000 2000
3050 300 8750 1550
Clusters 15 40 2 1 22 26 12
Applications 50 200 200 100 75 200 200
Licenses 25 40 4 20 20 40 25
VOSAP 1 4 5 1 2 1 4

Support Resource Requirements

Actual 25 15 3 3 10 23

Calculated 26 15 I 5 10 20

vendor, they started with all new equipment (Le., After spending some time working on this con-
no maintenance), they installed one current ver- cept and discussing it with a numberof managers
sion operating system, they had low user-support and staff at Project Athena, we feel quite com-
expectations initially, and theygot their initial in- fortable with this approach. Until more data are
stallation support,planning, and training from the available from otherinstallationsthatuse the
vendor. This is not a viable situation in the long Project Athena software, and until more installa-
run. tions have matured, this is our best estimate.

Verification of concept. We had the opportunity to An interesting extensionof this workwould be to

test our assumptions and modify our earlier esti- see if one can correctly predict the server con-
mates after the First International Athena Tech- figurations required tosupportthese environ-
nical Conference in April 1991. We obtained ments. Similar variables will be required, aug-
profiles of five Athena-like installations at univer- mented by others that are related to the type of
sities: North Carolina State University (NCSU), specialized centralservers,suchasKerberos,
University of Maryland (UMD), Iowa State Uni- Zephyr, Moira, etc., as well as customer require-
versity (ISU), and the University of California at ments for userfile space servers, backup require-
Santa Cruz (UCSC). In addition, we obtained data ments, and others.
from the University of Michigan (UMICH) in an
informal discussion at the Engineering Worksta-
tion Conference held in Boston in August 1991. Conclusion
Using these profiles, and making reasonable as-
sumptions in those cases where information was TheProjectAthenapartnershipbetween MIT,
incomplete, we constructed Table 1. Note that IBM, and DEC has provedto be very fruitful. It has
the formula works well for the larger installations given rise to new standards, such as the X Win-
with many clusters. North Carolina State Uni- dow System, and continues to make significant
versity has beenable to manage with a very small contributions to open systems architectures, in
staff up to now because they have single
a vendor particular to OSF’s DCE and DME. Until these ar-
installation, very few clusters, and new equip- chitectures have beenimplemented by the work-
ment. As they diversify equipment, distribute it to station manufacturers, many universities, com-
more locations, and expand their user services, panies,researchlaboratories, and government
more staff will likely be needed. Similarly, the agencies are looking to elements of the Athena
University of California, Santa Cruz, installation suite of network services tomanage their distrib-
is in the initial stages and presents an anomaly. uted systems installations.

562 ARFMAN AND RODEN IBM SYSTEMS JOURNAL, VOL31, NO 3, 1992

Our participation in Project Athena has demon- 10. M. E. Hodges, R. M. Sasnett, and M. S. Ackerman, “A
strated the value of joint projects and partner- Construction Set for Multimedia Applications,” IEEE,
37-43 (January 1989).
ships with universities in an open systems envi- 11. “Report Card on Technology: MIT Computer Network
ronment. Workstation installations from a single Offers Look at Future Education,” IBMMultimedia So-
vendor are the exception in the real world. It is lutions 4, No. 10, L. J. Baratto, Editor (October 1990).
therefore exceedingly important that we take ad- 12. “MIT Breaks NewGround with Interactive Multimedia,”
IBMMultimedia Solutions 4, No. 11, L. J. Baratto, Editor
vantage of the opportunities we have in the es- (November 1990).
tablishment of open standards. At the same time, 13. R. A. McKie, “Converting from RT to PSR: The Fun Is
it allows us to incorporate strategic architectures Just Beginning,” IBM Multimedia Solutions 4, No. 12
into our products at an early stage. (December 1990).

General references
Acknowledgments
C. R. Avril and R. L. Orcutt, “Athena: MIT’s Once and Fu-
We want to thank the management and staff of ture Distributed Computing Project,” Information Technol-
Project Athena for their thoughtful comments. ogy Quarterly (Fall 1990).
We particularly want to express our gratitude to G. A. Champine, MZT Project Athena, a Model for Distrib-
uted Campus Computing, Digital Press (1991).
Earl1 Murman, Ron Orcutt,CatherineAvril,
Naomi Schmidt, and Don Davis. Thanks also to
George Hebert of IBM for his feedback. Accepted for publication Januaty 20, 1992.
*Trademark orregistered trademark of International Business Josina M. Arfrnan IBM Corporation, Route 100, P.O. Box
Machines Corporation. 100, Somers, New York 10589. Dr. Arfman was IBM program
manager and MIT Associate Directorof Project Athena atthe
**Trademark or registered trademark of the Massachusetts Massachusetts Institute of Technology from November 1988
Institute of Technology, UNIX Systems Laboratories, Inc., until June 1991. Her currentIBM position is program manager
the Regents of the Universityof California at Berkeley, Digital in the Advanced Workstation Division. She joined the Re-
Equipment Corp., Apple Computer, Inc., the Open Software search Division in 1976, and has had a number of management
Foundation, Cray Research, Inc., Sun Microsystems, Inc., positions, including those in Corporate Ph.D. Recruiting and
Transarc Corp., or Parallax Graphics, Inc. Worldwide Internal Telecommunications Strategy. She has a
Ph.D. in X-ray crystallography.
Cited references
Peter RodenMassachusetts Institute of Technology,77Mas-
1. G. W. Treese, “Berkeley UNIX on 1000 Workstations: sachusettsAvenue, Cambridge, Massachusetts 02139. Mr.
AthenaChanges to 4.3 BSD,” USENIXAssociation Roden is the Manager of Distributed Systems in Information
(February 19G). Systems at MIT. Before the merger with Information Sys-
2. R. M. Needham and M. D. Schroeder, “Using Encrwtion tems, he was the Manager of Systems and Operations at Proj-
for Authentication in Large Networks of Compiters,” ect Athena. Prior to his coming to MIT, he was the Manager
Communications of the ACM 21, No. 12, 993-999 (De- of VMS Systems for Harvard University’s Faculty of Arts and
cember 1978). Sciences. He has a degree in computer science and is a mem-
3. J. G. Steiner, B. C. Neuman, and J. I. Schiller, “Ker- ber of ACM and IEEE.
beros, an Authentication Service for Open Network Sys-
tems,” USENIXAssociation (February 1988).
4. S. P. Dyer, “The Hesiod Name Server,” USENIXAs- Reprint Order No. G321-5486.
sociation (February 1988).
5. J. G. Steiner and D. E. Geer, Jr., “Network Services in
the Athena Environment,” USENIX Association (July
1988).
6 . C. A. DellaFera, M. W. Eichin, R. S. French, D. C.
Jedlinsky, J. T.Kohl, and W. E. Sommerfeld, “The
Zephyr Notification Service,” USENIXAssociation
(Winter 1988).
7.B. Anderson, T. Coppetto, D. E. Geer, Jr., and G. W.
Treese,“OLC: An On-Line Consulting System for
UNIX,” USENIX Association (February 1989).
8. M. A. Rosenstein, D. E. Geer, Jr., and P. J. Levine, “The
Athena
Service Management Systems (Moira),”
USENIX Association (February 1988).
9. K. Raeburn, J. Rochlis,W.Sommerfeld,and S. Zan-
arotti, “DISCUSS: An Electronic Conferencing System
for a Distributed ComputingEnvironment,” USENIXAs-
sociation (February 1989).

IBM SYSTEMS JOURNAL, VOL 31, NO 3, 1992 ARFMAN AND RODEN 563