simulation@ufl.edu (SIMULATION MODELING & ANALYSIS) (05/01/88)
Volume: 1, Issue: 10, Sat Apr 30 15:48:08 EDT 1988 +----------------+ | TODAY'S TOPICS | +----------------+ (1) Parallel Simulation (2) Battle Management/Simulation Experience (3) Simulating Parallel Algorithms ---------------------------------------------------------------------------- To: wagner@june.cs.washington.edu, simulation@ufl.edu Subject: RE: PDS paradigms Reply-To: aboulanger@bbn.com Date: Fri, 29 Apr 88 17:10:04 EDT From: aboulang@wilma.bbn.com Sender: aboulang@wilma.bbn.com wagner@june.cs.washington.edu writes: In other words, I think that interesting simulation paradigms are those in which processes execute ASYNCHRONOUSLY in real time, but SYNCHRONOUSLY in virtual time. Actually, I am experimenting with simulation paradigms which processes execute ASYNCHRONOUSLY in real time, but in general ASYNCHRONOUSLY in virtual time as well. (Synchronization may be an emergent property of the system as it self-organizes.) There has been work on this class of simulation/algorithms by various people. They are known as chaotic relaxation or asynchronous iterative methods. This was also studied by the Hearsay people in a more symbolic setting when they relaxed the synchronization locks on the blackboard, and the program seemed to still converge. These methods can be viewed a a step beyond virtual time. I am interested in a applying these methods to problems where the asynchronization actually HELPS the working of the program. Currently I am focusing on using the asynchronization as a noise source. (One example is a parallel diffusion limited aggregation program that uses the asynchronization among the processing elements as a source of noise.) What happens in this class of programs is that the dynamics of the parallel machine that the program is hosted on gets convolved with the dynamics of the program itself. Furthermore the dynamics of the machine can be separated out as a noise source in certain situations. The collective dynamics of the machine and program can be modeled as a time-delayed version of the original program dynamics, where the time delays follows a distribution that may or may not be well characterized. There has been work on the dynamics of time-delays systems, but as far as I know no mathematical studies of time-delay systems where there delay is variable and is characterized by a distribution. My work has been based on the Butterfly parallel processor. Cheers, Albert Boulanger aboulanger@bbn.com ---------------------------------------------------------------------------- Newsgroups: comp.simulation Subject: Simulating experience/Old message Date: 29 Apr 88 14:57:40 GMT Reply-To: king@rd1632.dayton.ncr.com (James King) Distribution: world Organization: R&D, NCR Corp., Dayton, Ohio Here is a message from a defunct news group from last year. I read this message and added a response. This may be of interest to the group and it may also cause some dicussion. Jim King j.a.king@dayton.ncr.com >Posted-From: The MITRE Corp., Bedford, MA >To: milsim@stl-host1.arpa >Subject: getting the discussion flowing... >Date: Tue, 23 Jun 87 12:44:11 EDT >From: jhs@mitre-bedford.arpa > >OK, here's a question for discussion that ought to get a response: > >A speaker at a MITRE-sponsored seminar (on the general topic of the efficienc >of the U.S. military procurement process) spent a few minutes talking about >the value of simulation in evaluating proposed weapons systems. He made >the comment that a very detailed simulation of a famous battle in history, >I believe it was Midway, showed our side losing, 0 for 200, in 200 runs. >In fact, we won. He said history shows that the reason we won was a small >number of "maverick" human decisions, which defied reason at the time but >proved to be right. Things like turning the fleet left instead of right >when all the rules in the book would have said turn right, because of a >"hunch" on the part of a commander. > >My question is this: In view of this experience, is it valid to depend on >Monte Carlo simulations for prognostications about how effective a system >will be? Should we be trying to incorporate more "worst case" or "blind luck >assumptions in our simulations to try to see how sensitive the system is to >these factors? Should we perhaps try to marry "expert system" technology wit >Monte Carlo methods, to come closer to reality? Does anybody have any other >suggestions on how to model the "nonstatistical luck" factor (for want of a >better term) in wargaming simulations? Can we learn anything useful about >military strategy from this area, such as the value of doing the unexpected, >and how to tell when to do so? > >There, that should provoke at least a LITTLE discussion! > >-John Sangster, MITRE. > I have been involved with battle management simulations in the past and have developed decision aids routines as a model of the human operator control in the simulation. The battle management simulation could exist as an entity operating under strict numeric computations and probabilities. It is when the human factor in the situation is modified that the greatest variability in simulation outcome occurs. Studies had even begun to model behavior patterns to input as responses to various battle situations. I have also proposed the use of analogical reasoning or modeling to be included into battle simulations. New or modeled situations can be evaluated and forecasted based on previous experience. The previous experience can be stored as "cases in memory and "remembered" by the system during simulation runs. I have prepared various reports and a paper on case-based reasoning under a general domain independent flavor. I'd be interested to learn about any work such as this presently under way. James A. King j.a.king@dayton.ncr.com (513)-445-1090 B ---------------------------------------------------------------------------- From: Tom Wisdom <gatech!hp-lsd!hpctdls!tsw@bikini.cis.ufl.edu> Date: Fri, 29 Apr 88 20:40:20 mdt To: fishwick@bikini.cis.ufl.edu Subject: Re: SIMULATION MODELING AND ANALYSIS Newsgroups: comp.simulation In-Reply-To: article <15241@uflorida.cis.ufl.EDU> of Thu, 28 Apr 88 15:40:16 MDT I am looking for references on simulating a multiprocessor running some parallel code. I am interested in studying the code speedup versus the number of processors. I am currently using a little discrete event simulation program called "smpl" written by M. H. MacDougall. My problem is that my simulation program is getting complex due to the synchronizations required between processors. Is there a standard approach to simulating parallel algorithms on a (simulated) multiprocessor? Thanks, Tom Wisdom Hewlett-Packard Colorado Telecommunications Division UNIX: hplabs!hp-lsd!hpctdlb!tsw P.O. Box 7050 SMTP: tsw@hpctdlb.HP.COM Colorado Springs, CO 80933 phone: 719-593-8700 x-737 +--------------------------+ | END OF SIMULATION DIGEST | +--------------------------+