Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!uflorida!gatech!hubcap!matloff
From: matloff@iris.ucdavis.edu (Norm Matloff)
Newsgroups: comp.parallel
Subject: Re: Distributed simulation (Discrete Event)
Message-ID: <5318@hubcap.clemson.edu>
Date: 28 Apr 89 12:11:34 GMT
Sender: fpst@hubcap.clemson.edu
Lines: 42
Approved: parallel@hubcap.clemson.edu

In article <5311@hubcap.clemson.edu> wen-king@csvax.caltech.edu (Wen-King Su) writes:
>In article <5276@hubcap.clemson.edu> matloff%mole.Berkeley.EDU@ucbvax.Berkeley.EDU (Norman Matloff) writes:
>>
><It has always seemed to me that [distributed] methods are inherently wasteful.
>>Each processor is going to have a substantial amount of idle time.
><
>>Moreover, it seems to me that the best method is also the simplest  --
><just have all P processors run the uniprocessor version of the program...

><This method gives 100% utilization for all processors...

>Since you are not worried about the elapse time of the simulation runs,
>which may is itself the most important consideration in some simulations,

I'm surprised to see you say this.  Am I missing something, maybe a
subtle distinction between elapsed time and some other measure of time?

I *am* worried about the time.  In fact, I even specified a time T
for a uniprocessor version, which it is hoped to reduce to T/P in a
P-processor system.

What has to be held constant in any comparison is accuracy.  I defined
the accuracy level to be whatever comes from running the simulation on
a uniprocessor for time T.  The problem with both kinds of distributed
simulations  --  the task-partitioned and replicated types  --  is that
in the former, there is less than 100% utilization, while in the latter,
there is initialization bias.  So in both cases, the desired accuracy 
needs more than T/P time, i.e. there is less than a factor-P speedup.
The only question is which is better.

>distributed simulation can be more efficient.  For the purpose of this
>discussion, let us first assume that:

>	1) Hardware used == the amount of memory involved.
>	2) Simulation requires the same amount of memory
>	   no matter how many processors are used.

As far as I know, memory is NOT a problem.  The major data structure is 
the event queue, and in typical applications this is not very long at all, 
maybe a dozen items.

  Norm