Web-Based Simulation: Revolution or Evolution?
Ernest H. Page Arnold Buss Paul A. Fishwick
The MITRE Corporation Operations Research Department Department of Computer
1820 Dolley Madison Blvd. Naval Postgraduate School and Information Science Engineering
McLean, VA 22102 Monterey, CA University of Florida
Gainesville, FL 32611
Kevin Healy Richard E. Nance Ray J. Paul
ThreadTec, Inc. Systems Research Center and Centre for Applied Simulation Modeling
P.O. Box 7 Department of Computer Science Brunel University
Chesterfield, MO 63017 Virginia Tech Uxbridge, Middlesex, UB8 3PH, UK
Blacksburg, VA 24061
The nature of the emerging field of web-based simulation is examined in terms of its relationship to
the fundamental aspects of simulation research and practice. The presentation, assuming a form of
debate, is based on a panel session held at the first International Conference on Web-Based Modeling
and Simulation which was sponsored by the Society for Computer Simulation during 11-14 January 1998
in San Diego, California. While no clear "winner" is evident in this debate, the issues raised here certainly
merit ongoing attention and contemplation.
Categories and Subject Descriptors: 1.6.5 [Simulation and Modeling]: Model Development
modeling methodologies, 1.6.8 [Simulation and Modeling]: Types of Simulation distributed
Additional Key Words and Phrases: Digital objects, distributed modeling, Java
The emergence of the world-wide web (WWW) has produced an environment within which many disciplines
are being re-evaluated in terms of their inherent approaches, techniques and philosophies. The disciplines
concerned with computer simulation are no exception to this phenomenon; the concept of ,-based sim-
ulation" has been introduced and is currently the subject of much interest to both simulation researchers
and simulation practitioners. As an area of scholarly endeavor, web-based simulation debuted as a 3-paper
session at the 1996 Winter Simulation Conference (WSC) and was, by far, the most well-attended session
within the modeling methodology track of that conference. This success was repeated at WSC 97, and in
January 1998 the first conference dedicated to the topic of web-based simulation was held as part of the
annual Society of Computer Simulation (SCS) Western Multiconference .
This paper stems from a panel convened for WEBSIM '98. The charter for the panel was to examine the
fundamental nature of web-based simulation and explore its relationship to the body of theory and practice in
simulation modeling methodology that has evolved over the past forty years . One goal for the panel was
to distill the essential and differentiating aspects of web-based simulation, if any, from amongst the mountains
hype that tend to surround the WWW. The central question was this: does web-based simulation represent
a revolutionary change or an evolutionary change? We posed the question because the nature of change
would seem to bear some relationship to the proper directions and focus of web-based simulation research
The panel composition was structured in an effort not only to portray all sides of the issues being
addressed but also, hopefully, to engender controversy and stimulate the participation of the audience.
Arnold Buss, Paul Fishwick and Kevin Healy are active in web-based simulation research and development.
Dick Nance and Ray Paul represent the "traditional" simulation modeling methodology community. Kevin
Healy presents a view from the commercial world, the rest of the panel hails from academe. Arnold, Dick,
Paul and Kevin provide a U.S. perspective. Ray serves as international representative.
The remainder of the paper is organized as follows. Section 2 establishes the framework for debate.
The panelists responses are captured in Section 3. We attempt to portray the essence of the dialogue that
occurred during the session through focus on a few key points of dispute in Section 4. Section 5 contains a
2 What are the Modeling Methodological Impacts of Web-Based
Simulation modeling methodology deals with the creation and manipulation of models over the lifetime
of their use. Motivated by the recognition that the manner in which a simulation model is conceived,
developed and used can have a significant impact on the ability of the model to achieve its objectives,
modeling methodology has been an active research area since the inception of digital computer simulation.
Over the past forty years the practice of simulation model creation has evolved from coding in general-
purpose languages, to model development in special-purpose simulation languages, to model design using
higher-level simulation model specification languages and formalisms, to comprehensive theories of simulation
modeling and holistic environment support for the modeling task. Thematic in much of the modeling
methodological work to date has been the recognition of Dijkstra's principle of the -I. p .. ... of concerns"
which argues for the separateness of specification and implementation . In many cases, this philosophy has
been tempered by the pragmatic observations of Swartout and Balzer , who observe that separation is a
worthy goal but not achievable in totality since any specification, S, may be viewed as an implementation
of some higher-order specification, S'.
Another argument in favor of the intertwining of specification and implementation is that technological
advancements may enable new approaches to accomplishing a task-approaches that were not even con-
ceivable prior to the advent of the technology. Consider, for example, the advent of the assembly line in
manufacturing. The potential of the WWW as such a technology push is cited in an article that describes
the application of the WWW within the manufacturing process , "Our initial experiments at putting en-
gineering, design and manufacturing services on the Web are so successful that we believe we should rethink
the traditional approaches and tools for coordinating large, distributed teams." With respect to simulation,
a similar revolution seems plausible. Web technology has the potential to significantly alter the ways in
which simulation models are developed (collaboratively, by composition), documented (dynamically, using
multimedia), analyzed (through open, widespread investigation) and executed (using massive distribution).
Is the web, in fact, such an elixir, demanding that we radically alter our modeling philosophies and
approaches? Or is web-based execution merely another implementation detail that can, and should, be
abstracted from the model development process?
The following sections contain responses from the panel members regarding the central question.
3.1 Web-Based Simulation Modeling
The explosion of computer networks have created an environment for computer modeling in general, and
simulation modeling in particular, that is revolutionary. In order to properly exploit these developments the
nature of modeling must change.
Simulation models have been traditionally monolithic in design. The advent of Object-Oriented Pro-
gramming has resulted in more elegantly designed monoliths. Simulation models for both industrial and
military applications have been mostly designed for models running on a single machine. For such mod-
els the network offers little. Using the full power of the network offers potentially substantial benefits to
modeling and simulation, but only if models are designed differently.
Use of up-to-date data by dynamically interacting with databases across the network speeds up the
modeling and decision-making cycle by an order of magnitude. The integration of computer models running
with systems has great potential for military analysis and training.
In nutshell, web-based simulation models must accommodate:
applications that expect to receive data across the network from a database that will be dynamically
applications that will expect to receive new classes and data unforeseen at the time the model was
applications using components that are loosely bound, rather than tightly coupled.
The Java programming language, together with the related cluster of Java Technologies, have substan-
tially extended the capabilities of program-level tasks. Java classes can open sockets across a network,
perform database queries, and encrypt data streams for secure transmission. New classes may be dynami-
cally incorporated as the program is running, thus enabling dynamic extensibility. Objects on one computer
may be serialized and sent to another, where thay are immediately incorporated into that computer's model.
Objects on another computer may be invoked through Remote Method Invocation.
The capabilities of programming languages have outstripped our knowledge of how best to write programs
exploiting these capabilities. Software design principles for procedural and even object-oriented programs
are well-known. It is not yet known how software should be designed using these tools. It is also not clear
how best to exploit the tremendous possibilities offered by the network.
3.2 Distributed Modeling Using the Web as an Infrastructure
(Paul A. Fishwick)
One of the most critical problems in the field of computer simulation today is the lack of published models
and physical objects within a medium-such as the World Wide Web-allowing such distribution. The web
represents the future of information sharing and exchange, and yet it is used primarily for the publication
of documents since the web adopts a "document/desktop metaphor" for knowledge. In the near future, we
envision an "object metaphor" where a document is one type of object. A web predicated on digital objects
is much more flexible and requires a knowledge in how to model physical phenomena at many different scales
If a scientist or engineer (i.e., model author) works on a model, places the model inside objects, and
constructs a working simulation, this work occurs most often within a vacuum. Consider a scenario involving
an internal combustion engine in an automobile, where the engine is the physical object to be simulated.
The model author's task is to simulate the engine given that a new engineering method, involving a change
in fuel injection for example, is to be tested. By testing the digital engine and fuel injection system using
simulation, the author can determine the potential shortfalls and benefits of the new technique. This task
is a worthwhile one for simulation, and simulation as a field has demonstrated its utility for objects such as
Let's analyze the problems inherent in this example. There is no particular location that will help the
author to create the geometry of the engine and its dynamics. Moreover, if the model author seeks reusable
components on the web, who is to ensure the quality or accuracy of these components? It may be that
other employees of the company have made similar engine models in the past, and that these models may
be partially reused. If this is the case, the model author is fortunate, but even if such a company-internal
model exists, it may not be represented in "model form." There may be other model authors who have
already constructed pieces that our model author could use, but there if there is no reuse and no standard
mechanism for publishing the model or engine object, then this is all for naught. The model author may also
be concerned with creating a fast simulation. While algorithms for speeding simulations are important, by
solving the reusability problem, we also partially solve the speed problem since published quality models of
engines will battle in the marketplace for digital parts, and the best engine models and testing environments
involving very fast and efficient simulation algorithms-will win out in the end. Therefore, the problem of
reuse of engine objects and components lies at the heart of the simulationist's dilemma. Fast, efficient
and quality models could be available at some point in the future, but today there is no infrastructure or
agreed-upon standards. for true digital object engineering.
What if the model author of the engine creates a digital engine that operates differently than the actual
one? The automobile company could provide full access to an invalid model. We must have quality control
measures in place to help us with this situation. The physical metaphor provides some help. Many consumer
groups and institutions exist to protect consumers from bad products. Digital products will require similar
groups and testing procedures. If a company knowingly markets a bad digital product, they will ultimately
pay for this error in the marketplace. The digital object must be treated with the same level of quality control
as the physical counterpart. In some cases, a company might make a mistake in production and a part or
entire vehicle must be recalled. This type of recall is made easier with the digital product. It behooves the
model authors to create valid, quality objects. It may be that anyone can publish a digital object but this
is true of physical objects as well. The situation is somewhat more acute with a digital automobile since to
create an automobile in the first place, one must have invested a fair amount of time and resources; however,
a digital engine could be created by the neighbor down the street. One must learn to trust certain sources
more than others based on past performance of prior digital objects. Also, we must have ways of verifying
our sources, developers and producers with methods such as digital signatures, watermarks and encryption.
3.3 Simulation Modeling Methodology and the WWW
(Kevin J. Healy)
The World Wide Web was conceived as a set of simple Internet-based client/server protocols for transferring
and rendering documents of a primarily textual nature. What distinguished the Web's mode of communicat-
ing information from other Internet-based tools that preceded it (e.g. electronic mail, electronic file transfer
via ftp, and network newsgroups) was the provision for embedding hyperlinks that allowed users to easily
navigate between related documents. The hyperlinking scheme allowed content providers to organize and
present information in a natural hierarchical fashion. It also served to insulate users from the tedious details
involved in identifying and retrieving a particular document. Since the development and rapid widespread
adoption of these conventions, they have been extended and integrated with other new related technologies
that provide for the delivery of content that is much more dynamic in nature. The most important of these
related developments has been the introduction and rapid widespread adoption of the Java programming
language as a standard for Internet-based computation.
The integration of the Web and Java represents a technological advancement that enables a fundamen-
tally new approach to simulation modeling, one that makes possible the development of environments with
coherent Web-based support for collaborative model development, dynamic multimedia-based documenta-
tion, as well as open widespread execution and investigative analysis of models. A key aspect to the approach
is the role the Java language plays in both the specification and implementation of the model.
The evolution to high-level model specification languages and formalisms has been motivated by the
desire to make simulation more accessible by eliminating the programming burden. However, such systems
are often difficult to modify or extend because of an imposed separation between the specification system
and its implementation. This can lead to models that poorly mirror system behavior and have no potential
for distribution and reuse within an enterprise. The Java language is ideally suited to implementing an
advanced simulation architecture whose features are readily accessible at the programming language level,
special purpose simulation language level, and high-level model specifications.
Specifically, key features like the well-designed object-oriented nature of Java and native support for mul-
tithreaded execution allow special purpose simulation modeling features to be incorporated directly into the
Java language in a natural way so that the underlying modeling and programming languages are the same.
These relatively low-level but powerful modeling capabilities can in-turn be used to implement higher-level
model specification systems via the JavaBeans component development model. The simple programming
conventions that constitute JavaBeans allow Java-based software components to be assembled visually into
applications using any of a growing number of sophisticated graphical programming environments including
Symantec's Visual Cafe, Microsoft's J++, IBM's Visual Age, Sun's Java Workshop, Borland's Jbuilder, and
Lotus's BeanMachine. When visually assembling predefined simulation modeling components, no program-
ming is required; however, when necessary, the user has access to the underlying code and full power and
flexibility of the Java programming language. What's more, any Java environment can be used for model
building and debugging. The modeling language capabilities and predefined component assembly capabilities
can also be used in isolation or in combination to produce high-level standalone simulation applications that
users interact with in predefined ways.
The hardware and operating system independent design of Java facilitates collaboration by allowing
modelers to share language level or component level models independent of where they were developed. The
documentation and deployment of modeling tools and end-user applications via the Web also serves to make
open and widespread both the development and investigative analysis of models.
This vision of Web-based simulation is the motivation behind Thread Technologies' design of s I/
a general purpose simulation language based on the Java programming language. Silk merges familiar
process-oriented modeling structures with powerful object-oriented language features in an intelligent de-
sign that encourages model simplicity and reusability through the development and the visual assembly
of Silk modeling components in JavaBeans-based graphical software environments. More generally, Silk's
openly extensible, scalable, and platform independent design represents the type of approach that is essen-
tial to keeping simulation modeling on track with other revolutionary changes taking place in Internet-based
3.4 Simulation Modeling Methodology in the Wonderfully Webbed World
(Richard E. Nance)
While modeling methodology has been with us since the inception of simulation, it remained indistinguish-
able from programming throughout the first two decades. Nevertheless, a few early researchers abstracted
beyond the executable form to search for more significant semantic revelations. Lackner and Kribs  and
Kiviat  are prominent examples, but Tocher's  wheel charts to assist in model specification and the
IFIP proceedings on simulation programming languages  show that interest was widespread. Efforts to
derive a theory of simulation  generated interest in model representation in the 1970s. The latter part of
the decade ushered in the first specific focus on modeling methodology (model life cycle, model specification
languages, the DELTA project) . With the 1980s came the vision of model development environments 
that are now a commercial reality. Is the subject of this panel session presaging the next major transition
in simulation model development?
3.4.1 Modeling Methodology
Since "methodology" is both over-used and misused, a definitional explanation in this context is appropriate.
Methodology, following the view of Arthur et al. [1, p. 4], should:
organize and structure the tasks comprising the effort to achieve global objectives,
include methods and techniques for accomplishing individual tasks (within the framework of global
prescribe an in which certain classes of decisions are made and the ways of making those decisions
leading to desired objectives.
Key in the attainment of the objectives are the principles that form the foundational support of a method-
3.4.2 Influence of the Web
If the world wide web is to effect major changes in modeling methodology, then it must alter or abolish
existing principles or introduce new principles. At this juncture the capability of the web to influence the
technology of model building, model execution and model sharing is clear, and the degree of change appears
significant. However, that the potential for influence extends into the principles -the foundational core -is
3.5 Web-Based Simulation: Whither We Wander?
(Ray J. Paul)
This panel contribution will discuss a variety of new technologies for software development and ways of
working that will have an unpredictable effect on the future of simulation modelling.
3.5.1 Multi-media/Synthetic Environments
The ability to access multi-media on the web clearly introduces greater potential for the use of videos of
problem scenarios, for interaction with stake-holders situated at remote locations (for example, when the
running model hits an unknown combination of circumstances, an expert stake-holder might be able to
determine the successful rules for advance) and sound. For example, on a recent visit made to a Hong
Kong container terminal, I was shown a television control centre, computer-based, which had 1111' video
coverage of the terminal. Whilst its purpose was clearly for security and safety, it requires little imagination
to visualise how a simulation of the terminal operation could call up the appropriate video camera when
problem discussants get to the point of a simulation run where clarification is desired. I think that the
rush to join the much-hyped band-wagon of Synthetic Environments, driven by technical extravagance and
financial greed, is in great danger of neglecting or even forgetting those major simulation issues of ongoing
concern over the years. These are the so far intractable problems of verification and validation. The current
enthusiasm for Synthetic Environments is therefore in danger of creating more expensive mistakes to the
detriment of the reputation of simulationists, analysts and operations researchers in general.
3.5.2 Natural Born Webbers
A large proportion of the current generation of students entering higher education in the developed countries
are already familiar with the pastime of browsing the Web and playing computer games. Both of these
activities might loosely be depicted as approaches based on "suck it and -. Browsing and adventure
games encourage the participant to try out alternatives with rapid feed-back, avoiding the need to analyse
a problem with a view to deriving the result.
Such web users, in order to use simulation, need and desire development tools that allow for fast model
building and quick and easy experimentation. Furthermore, such web users should have a natural affinity to
the use of simulation models as a problem understanding approach [15, 16]. Web-enabled simulation analysts
will be opposed to classical software engineering approaches and methodologies. They will be seeking tools
that will enable them to assemble rather than build a model. Some feel for the change of "culture" that we
can expect from future generations of computer users can be gauged from a recent experience of mine on
a visit to Taipei (Taiwan). A class of school children were using the local university's multi-media lab. A
ten year old schoolboy was typing in HTML codes faster than I can and dynamically checking it by running
a rather impressive text/video/sound demonstration system. The boy could not speak, read or write any
English, everything was symbolic to him.
3.5.3 New Software Technologies
Some have predicted that the software industry will be divided into component factories where powerful and
general components will be built and tested, and into component assembly shops where these components
will be assembled into flexible business solutions. Such component based development, if it occurs, might
give significant gains in productivity and quality as well as known obvious benefits to web-based software
Java is now so ubiquitous that it might appear unnecessary to comment on it. For completeness the reader
is reminded that simulation models in Java can be made widely available; an applet can be retrieved and
run and does not have to be ported to a different platform or even recompiled or relinked; there is a high
degree of dynamism because Java applets run on a browser; Java built-in threads make it easier to implement
simulation following the process interactive paradigms; Java has built in supports for providing sophisticated
animations and Java is smaller, cleaner, safer and easier to learn than C++, allegedly.
For me, the aforegoing indicates a world of dynamic change, which I welcome, but where it is all going is a
matter of conjecture that will be colored more by prejudice and opinion than evidence.
In this section the authors respond to the points made in the previous section.
4.1 Ray Paul's Comments
Regarding the positions of Arnold Buss and Kevin Healy, it is arguable that Java is so good. We have
experience of platform dependence, and of course the rate of enhanced releases which are not downward
compatible outdates software rapidly. On the other hand, such fast adaptation of the language might
encourage improved methods of release compatibility, to the benefit of the industry at large.
Regarding Paul Fishwick's position, it is arguable that quality control is necessary for software re-use.
The traditional methods of building large models, which takes much time and money, and which in itself
then leads to an expectation of repeated use, demand some sort of quality assurance. When it takes so long
to get an answerss, it is a bit limp to also admit that the model may be indeterminately wrong! However,
if we can "glue" bits together fast and experimentally (Ray's crystal ball in action here), then maybe the
emphasis will shift dramatically from "is the model correct?" to "is the analysis, albeit with unproven
software, acceptable given the large experimentation that swift modelling has enabled us to carry out in a
short space of time?" In other words, the search space has been dramatically reduced not by accuracy (the
old way), but by massive and rapid search conducted by an empowered analyst (the new way).
Regarding Richard Nance's position, maybe our current principles are inappropriate for a web-based
world. I have already argued some of this in the previous paragraph. Here I go further. We are in a period
of rapid technical change (though some authors claim this will come to an end and life will settle down
again -see  ). Every attempt we make to use these technological advances adds to the opening up of new
opportunities to make change. This is particularly noticeable in business, where new companies are emerging
fast, old ones sinking daily, mergers, acquisitions, takeovers, etc. are prevalent. Even in the military sphere,
the nature of the task to be faced changes quickly (war, peace-keeping, policing, training allies, reassessing
threat as the political world moves on and so forth). Analysis needs to be fast, else the problem has moved
on anyway. Methods that produce ballpark estimates quickly, enhanced with more accurate methods if
time allows, are or will be the order of the day. Principles based on output analysis, rather than modelling
analysis, are likely to be more appropriate. If the traditional analytical and academic communities try to
maintain current principles, they will become historians, worthy of a footnote about Luddite Neanderthals
in the next Millennium history.
4.2 Arnold Buss's Comments
Regarding Ray Paul's comments, he has indeed brought up some thought-provoking issues. With regard
to Java, although it is good to be skeptical, it is clear at this point that the only thing that will derail its
achieving true platform independence is willful destruction, to wit, Microsoft's attempts to make it Windows-
specific. There is, in my opinion, simply too large a critical mass of developers and companies who are getting
on board for this to happen.
Moreover, I believe that the Java component infrastructure (JavaBeans) will be precisely the platform
on which to assemble large models from smaller components, so that the entire monolith does not have to be
designed in one piece. I believe that component-based design will supplant 00 design in a major way in the
near future, in part fueled by network-based computing. On the network, you must be component-oriented
or the thing is just too unwieldy. Designing distributed models in a reasonable manner pretty much forces
you into components. The design issues focus more on responsibilities and interoperability rather than class
hierarchies, as in traditional 00 design.
There is a somewhat subtle aspect of the Java language that turns out to be the real winner for component-
based design, namely interfaces (vice classes). Interfaces enable components to interoperate and pass mes-
sages without having to know the precise class or class hierarchy of each other. The interface is simply a
contract to implement certain methods, so they may be invoked with compiler-safe impunity. Interfaces
allow you to replace one object with another of an entirely different class with no necessary implied "is-a"
The second really important element is enabled by interfaces: communication via events. Interfaces allow
you to define a small handful of event sources and event listeners that can provide communications between
objects that is much more flexible than ordinary method invocation. One object will register its interest in
another's events (or, more likely, be registered by a third party). Whenever the event source's state changes
to trigger an event, all objects listening are notified. The key is that neither event sources nor listeners
need to be i- -. that any of this is happening. Objects can register and un-register their interest as the
program evolves. Event communication is a powerful means of implementing distributed models. Remote
objects may easily register as listeners by using a remote mechanism (RMI, CORBA, etc.) and the event
sources need not know (or care).
I have enhanced Simkit to incorporate this kind of messaging, and am currently working with a student to
extend it further. For example, we have generic entities that are nothing more than containers. F,,. I1. .11 ilil -r
is put on these entities by creating and adding components. To enable movement, for example, a Mover
entity is thrown in. The kind of movement possible is entirely determined by the type of Mover. Add a
sensor and you can detect other things (depending, of course, on the kind of sensor). If a Mover and a
Sensor are put together in a container, then the movement is governed by the Mover. Basically, this is an
extremely flexible type of composition. It is difficult to express in standard 00 notation, since neither the
Mover nor the Sensor are instance variables. Besides, Booch/UML diagrams just tend to confuse matters
in my opinion. The point is that a generic component-based methodology needs to be developed for such
modeling. Java is a perfect vehicle for doing just that.
4.3 Richard Nance's Comments
Regarding Ray Paul's comments, I do not accept the claim that "it is a bit limp to also admit that the
model may be indeterminately wrong." A model is not reality and only a fool insists that a model be error-
free (the same person who wants a world with no accidents). How do you propose to answer the shifted
question above: "Is the analysis ... acceptable?" I see your only recourse as an after-the-fact conclusion,
which advances us to the stage of relying on prophets-why bother with the unproven software, etc? Having
returned us to the technology of 2000 years ago, what next is to be offered? How about a roulette wheel
with labeled outcomes?
Since the good Dr. Paul does not provide quotation marks or a page number, I assume that these
sentiments are not those of Fernandez-Armesto but his own. My reaction is that I do not live in the fast
lane that engulfs Dr. Paul. My long-time technical sponsor, the US Navy, is working now and for some
three years prior on the design of the next destroyer (2003). The models and simulations used in this task
are time-consuming to develop and the analysis is conducted over years, not days, hours or seconds. The
hull will be in service for some 30 years and undergo three major overhauls all of which will require parts
of the existing models and still others that will be developed, again perhaps in months, but certainly not in
I do not think our modeling methodology principles have been altered at all by the web. The capabilities
of the tools based on the principles have changed and are changing, but that is the way technical progress
My thanks to the good Dr. Paul for his agitating expressions of these misguided views. May he never fly
on an aircraft developed with his analysis/prophet approach to decision making.
4.4 Paul Fishwick's Comments
F- .,,i. on the panel makes valid points. There is a need to tie together some of the views to make a
whole. At the same time, I'll express my own perspective on Ii. i. it all is going." Ernie Page's reference
of Dijkstra's principle is one where we must separate specification from implementation. In general, this
separation is one where we talk of I. I of model." A piece of code, a mathematical expression, a Petri
network, and a 3D cylinder are all examples of models. We may choose one specific model to represent some
aspect of the system that we are studying. The code may represent the same dynamics as the Petri net,
while the cylinder may represent either an abstraction of the system shape or, perhaps, a state of the system.
The interpretation that we foster is the essence of modeling. Modeling is an art in this respect.
Arnie Buss and Kevin Healy speak kindly of Java. Java does show promise for its intended function: a
computer language meant to migrate over a large area network to promote distributed computing. At the
same time, Java is a textual computer language so its primary purpose is to represent dynamics at a fairly
low level of abstraction. I'll submit that source code written in Java is a model, and that one can "think in
I ,- about system behavior. On the other hand, many people will not find this metaphor as appealing as
one that is visual and graphical. Models must serve the user's view and way of thinking. There is no one
correct modeling language. Ultimately, models are shared metaphors. If I am part of the Petri network or
System Dynamics community, I think in these specific icons. The models color my thinking about dynamics.
If we can all agree that we have many modeling types or languages, and that we can form translations
among models (from Petri networks to Java, for example), then all forms of modeling become germane to
the discussion. With Java at the lowest level of translation, our task of distributed execution of models is
enhanced, and so research and development of Java is good for web-based modeling and simulation. Let's
just keep in mind that Java is one of many nodes in a vast modeling network with models as nodes and
translations as arcs. I know very few scientists or engineers who would prefer Java over their pet modeling
Dick Nance and Ray Paul speak of two opposite poles in terms of quality in modeling. If I attend
an art exhibition and buy a modern sculpture made of electronic home appliances-such as toasters, can
openers, and mixers-I will most likely not use this artwork for engineering purposes. When special effects
companies in Hollywood create models of the Titanic and of New York City, their objective is to foster
entertainment and not to create statistically valid behavior. Therefore, both views are supported. Quality
must be maintained where it is required, and to the degree that it is required depending on overall objectives
of the simulation. There is nothing wrong with the Taiwanese schoolboy (Paul's example) who grabs objects
left and right to create a new experience. Some of these objects, like the toaster in the sculpture, may be
based on high resolution models (both structural and dynamic). It is the way in which the objects are used
that determines the outcome, and all outcomes are fair game.
Luckily, the future is bright for simulation and web-based modeling and simulation. Imagine the Tai-
wanese schoolboy unchained from abstract languages such as HTML. Instead, a new world-wide marketplace
of digital objects yields the digital equivalent of everything you see around you. The web is no longer fettered
with documents. Documents are but one kind of physical object. The schoolboy will be creating complex
games and simulations for his friends who will later join him in a multiplayer extravaganza. Meanwhile,
the Navy is testing out a new class of submarine using objects delivered by its contractors. This delivery
occurs well before the physical submarine components are manufactured. Some of the Navy objects will
be the same used by the schoolboy just as the toaster can be used in more than one way. The objectives
of the Navy and schoolboy are different, but the digital object marketplace is common to both of them. I
think that Dick Nance is right about a change in modeling methodology. It is happening now and web-based
simulation is the catalyst. The purpose of physical objects is to achieve a singular objective, but the global
objective is left open to the end-user. This is a departure from Arthur et al.  We should not limit our
models by global objectives. Objectives and models are orthogonal. I use the web to locate objects, and I
use these objects to create models of every variety. Like the manufacturer of the toaster, I create the best
digital toaster possible and let the consumer make the choice as to its utility. I would not be at all surprised
to go to Taiwan in ten years and find the schoolboy playing a "multi-player deathmatch" inside the confines
of a greatly enlarged toaster within a Dali-esque landscape. Meanwhile, the Navy is modeling the high-level
dynamics of a towed-array sonar using a circuit of light bulbs from General Electric's web site, with bulbs
The era of web is certainly upon us. There seems to be no escaping that fact. The confluence of the web and
simulation offers an opportunity to change the way we approach modeling. How much should we embrace
such change? The panelists disagree on this point. If the world of digital objects appears-and if publishing
models on the web becomes profitable, digital objects will proliferate-will the modeling process become
enhanced or impaired? Certainly the act of model construction would be simpler-assuming sufficiently
powerful search engines. It should be much easier to p'l",' models together than to build models from
scratch. But then what? How will models be validated in this environment? Unless the open source
movement achieves ubiquity, model validation may be one big exercise in black-box testing. In areas where
validation is critical, this situation can only represent a step backwards.
But perhaps an engineering analogy is useful here. No one would argue, for example, that bridges are a
bad idea. Although occasionally failures do occur (and such failures can be catastrophic) for the most part
bridges are engineered for safety. Where possible, pathological situations are considered and accounted for
in bridge design. Worst-case capacities (and then some) are accommodated. The opportunity to misapply
the science and mathematics that support bridge design exists, but the engineering profession actively seeks
to limit such opportunities.
Technology marches on. Modeling is central to technological advancement. But advancing technology
impacts the modeling process as well. As simulation becomes a desktop commodity, it will be available to
masses. This ubiquity is a mixed blessing. Having access to such a powerful problem-solving technique is
potentially quite valuable. On the other hand, to the untrained user-a user with a what-you-see-is-what-
you-get perspective-the potential to misapply the technique is great. As responsible engineers of the future,
should those enabling the web-based simulation revolution also shepherd the safety of the technique?
 Arthur, J.D., R.E. Nance and S.M. Henry. 1986. "A Procedural Approach to Evaluating Software
Development Methodologies and Associated Products." SRC-87-007, Systems Research Center, Virginia
Tech, Blacksburg, VA.
 Buxton, J.(ed). 1968. Proceedings of the IFIP Working Conference on Simulation Programming Lan-
 Cubert, R. and Fishwick, P.A. 1997. :,lOOSE: An Object-Oriented Multimodeling and Simulation
Application It ..... ..i1: accepted for Simulation, July 1997. Further information on MOOSE can be
found in http://www.cise.ufl.edu/ ~fishwick/moose.html
 Dijkstra, E.W. 1976. A Discipline of Programming. Prentice-Hall, Englewood Cliffs, NJ.
 Erkes, J.W.; Kenny, K.B.; Lewis, J.W.; Sarachan, B.D.; Sobolewski, M.W; and R.N. Sum, Jr. 1996.
"Implementing Shared Manufacturing Services on the World-Wide Web." Communications of the ACM,
39, no. 2 (Feb.):34-45.
 Fernandez-Armesto, F. 1995. Millennium: A History of the Last Thousand Years. Touchstone, Inc. New
 Fishwick, P.A. 1995. Simulation Model Design and Execution, Prentice-Hall, Englewood Cliffs, NJ..
 Fishwick, P.A. 1996. "Web-Based Simulation: Some Personal Observations" In: Proceedings of the 1996
Winter Simulation Conference, (San Diego, CA, Dec. 8-11). Association for Computing Machinery, New
 Fishwick, P.A., Hill, D.R.C. and Smith, R., Eds. Proceedings of the 1998 International Conference on
Web-Based Modeling and Simulation, SCS Simulation Series 30(1), San Diego, CA, 10-14 January 1998.
 Kiviat, P.J. 1967. "Digital Computer Simulation: Modeling Concepts." RAND Memo RM-5883-PR,
Santa Monica, CA.
 Lackner, M.R. and P. Kribs. 1964. Ini I..l. I... i to a Calculus of Change." System Development Corp.,
 Nance, R.E. 1979. ;..... I Representation in Discrete Event Simulation: Prospects for Developing
Documentation Standards." In Current Issues in Computer Simulation, N. Adam and A. Dogramaci
(eds), Academic Press, 83-97.
 Nance, R.E. 1983. "A Tutorial View of Simulation Model Development." In: Proceedings of the 1983
Winter Simulation Conference, (Arlington,VA, Dec 12-14). IEEE, Piscataway, NJ, 325-331.
 Page, E.H., Buss, A., Fishwick, P.A., Healy, K.J., Nance, R.E. and Paul, R.J. 1998. "The Modeling
Methodological Impacts of Web-Based Simulation," In: Proceedings of the 1998 SCS International
Conference on Web-Based Modeling and Simulation, pp. 123-128, San Diego, CA, 11-14 January.
 Paul R.J and D.W. Balmer 1993. Simulation Modelling. Lund, Sweden: Chartwell-Bratt Student-Text
 Paul R.J. and V. Hlupic 1994. "The CA'.. I Environment Revisited." In: Proceedings of the 1994 Win-
ter Simulation Conference, (J.D. Tew, S. Manivannan, D.A. Sadowski and A.F. Seila, Eds.) (11-14
December 1994, Orlando). Association for Computing Machinery, New York, 641-648.
 Swartout W.; and R. Balzer. 1982. "On the Inevitable Intertwining of Specification and Implementa-
tion." Communications of the ACM, 25, no. 7(July):438-440.
 Tocher, K.D.T. 1966. "Some Techniques of Model Building." In: Proceedings, IBM Scientific Symposium
on Simulation Models and Gaming, pp. 119-155, White Plains, NY.
 Zeigler, B.P. 1976. Theory of Modelling and Simulation, Wiley, New York.