Ventana software support forum

Just to add to that, to my understanding:

1) Level of detail: Systems dynamics looks at systems at a very high level
so is more suited to strategic analysis. Discrete event looks at the nitty
gritty detailed analysis and is more suited to process re-engineering
types problems.

2) Indicitive vs predictive modelling: Systems dynamics is indicitive , ie
helps us understand the direction and magnitude of effects (ie , where in
the system do we need to make the changes) , whereas discrete event is
predictive (ie how many resources do we need to make get X level of
throughout).

3) Continous vs discrete: Systems dynamics analysis is continous in time
(tends to use deterministic analysis) , whereas discrete event deals with
analysis in specific time periods (and tends to use more stochastic
analysis).

At Accenture, we have used both as complementary tools. SD to look at the
high level problem and identify areas which need more detailed analysis. We
then use discrete event modelling tools to analyse (and predict) the
specific areas we want to look at. You can link the two together and get
them to exchange data using Excel.

The new version of powersim (i think 2001) has now got discrete event
modelling functionality, which means that you can do both types of
modelling in powersim and don;t need to link 2 different pieces of software
via Excel.

Hope that helps !!

Rohita
From: rohita.singh@accenture.com

I think the difference between system dynamics folks and discrete event
folks is the following:

1) The modeling **techniques** used by the two different groups are usually
different at two levels: (1) The modelers way of representing systems is
different and (2) the underlying simulators algorithms are also different.
Each technique is well tuned to the purpose of the group that uses it.
Nonetheless you could use a Discrete Event approach to do system dynamics
and you could use a system dynamics modeling approach to do what discrete
event folks usually do.

2) The most important difference is the **purpose** of the modeling. The
discrete event folks are trying to find how many resources you need (e.g.
how many trucks) and how to arrange the resources in order to avoid
bottlenecks (excessive lines, waiting times, or inventories). The SD folks
are trying to figure out how to change decision making (e.g. having people
respond faster or slower or use different information) and sometimes how to
change the physical structure (e.g. physical shipping delay time) so that
inventories, sales, production, etc. wont oscillate or show some other
undesirable pattern of behavior.

As part of a larger research project, our group is exploring the combination
of SD with discrete event techniques. We dont have anything to report at
the moment.

Regards,
Jim Hines
jhines@mit.edu

Hans Scholl has written a good paper comparing system dynamics with agent-based
modeling where he shows how the two modeling techniques fit together.

The article contains good, concise explanations of sd and agent-based modeling
along with their relative strengths and weaknesses.

You can find a pdf format version of the article at:
http://dlib.computer.org/conferen/hicss ... 813003.pdf

Michael
_______________________________________

Michael Bean
Forio Business Simulations

mbean@forio.com
www.forio.com

Rohita Singh gives us some additional contrasts between SD and discrete
event simulation. I have one small clarification, though on her point 3,
which deals with discrete vs. continuous.

The contrast here is not between discrete vs. continuous time, but rather
between whether events happen at a single point in time or are smeared over
a period of time. An event in a discrete event model can happen at any point
in time. However, when it happens, it happens all at once, in an instant.
In traditional SD modeling events can also happen at any point in time, but
when they happen they tend to happen "gradually" over a period of time.

The implementation of the two techniques (SD vs. DEM) usually involves
different implementation of the "clock" that keeps track of time. In system
dynamics simulations, the clock is really just a level with a single inflow
of time. The equation in DYNAMO would be
Time.k = Time.j + dt

Discrete event-type simulations **could** use a system-dynamics-like clock,
but usually dont. Instead, the discrete event clock usually makes use of
the discrete event simulations "event queue" which is simply an ordered
list of upcoming events, each of which has a time-stamp for when it is
slated to occur. The time-gap between any two events is not constant. For
example, the events in a queue could have time stamps: 2.3456, 7.28106,
7.6987, 27.9010234. The queue is ordered from soonest to latest. The
program takes the first event off the queue, sets time equal to that events
time-stamp, and executes the event. Then the program takes the next event
off the queue, sets time equal to the enw events time-stamp, and executes
that event. ETC. The events can have ANY time stamp. Given this
representation you could say that time proceeds in discrete, though
unequal, jumps. But really the idea is that time is continuous, and at any
given instant either an event happens happens or nothing happens. Discrete
event implementations COULD use the SD type clock, but instead usually try
for increased efficiency (and precision) by skipping over periods when
nothing happens.

Jim
From: "Jim Hines" <jhines@MIT.EDU>