Posted by Bill Harris <
bill_harris@facilitatedsystems.com>
""SDMAIL George Richardson"" <
gpr@albany.edu> writes:
> Beyond the questions of software and integration of approaches are
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
George,
With all the suggestions of commercial software that does both DE and SD
simulation, I wanted to mention one of the early GPL'd packages that
does all of this and more: Paul Fishwick's SimPack. It is a C language
system that supported his notion of multi-modeling, in which one might
have models, sub-models, and sub-sub-models (and so forth) where each
model might be built using a different methodology. So you could have,
for example, a top-level finite state automaton that, depending upon the
time of day and day of the week, could invoke one of several SD models
that would describe the behavior of shoppers in a grocery store. One
might inquire about the purpose of such a model and why it might be more
attractive than other alternatives, but that's the sort of thing it can
do.
In the FAQ of the once active comp.simulation USENET group, SimPack was
listed as one of the basic tools for a simulationist's toolkit. Paul
wrote a book about SimPack called Simulation Process Design and
Execution. I wrote an article for The Systems Thinker (System Dynamics
on a Shoestring) that told how to do SD simulation using SimPack
(
http://www.thesystemsthinker.com/nldata ... ItemID=590).
> > A discrete event approach seems to need to assume that nothing happens
> > between events. (Is that a correct characterization of the approach?)
I'd not classify myself as a devotee of DE simulation, but I have used
it. I'd perhaps modify your sentence to, ""A discrete event approach
seems to need to assume that nothing _of current interest_ happens
between events."" (Horses for courses and all that).
Does that help? I suspect it may be a fairer portrayal, but you might
still question whether something in the time between events /should/ be
of current interest.
Not having used DE simulation in a while, I'm curious: when people here
speak of the need for the DE methodology, to what are they referring?
What do they understand as DE simulatino?
Is it the ability to track characteristics of individual entities
without using coflows (i.e., discrete entities in the conserved
flows)?
Or is it more the focus on discrete time: the ability, for example, to
model queueing behavior stochastically?
Or do DE models communicate better to certain groups less familiar
with SD and uncomfortable with the level of aggregation we tend to
use?
Bill
- --
Bill Harris
Posted by Bill Harris <
bill_harris@facilitatedsystems.com>
posting date Tue, 08 Jan 2008 09:22:29 -0800
_______________________________________________