Re: Maximum size

[Paul E Johnson]

I don't think there is a good answer to this question.  As you say
yourself, this is not a very precise question, since the size/complexity
of the agents influences the size of the project.  But I asked this
question too a while ago, so I'll just tell what I see.

I've built one model that I run with about 1 million simple agents and
50 complicated ones.  That uses about 140 meg of RAM, as I recall, after
about 10 minutes.  I would estimate each of the simple agents has only
five or six doubles, five ints, and a half dozen pointers allocated to
it.  

In a C book I recently reread a passage about how variables get packed
into RAM and it can make a big difference how you declare your
variables.  I think it said declare smallest varibles first (bool,
unsigned, int, long, float, double,...) I did some checking and in the
model with 1 million agents, it saved a couple of megabytes of memory.
Or maybe I was fooling myself.

I have another model that puts agents in a square grid, as in a cellular
automata.  Each agent makes some fairly complicated calculations on the
basis of its neighbors, and I find that a grid of size 25x25 is about as
high as I want to go before the whole model becomes intolerably slow.  

For exploratory purposes, I find that, unless I can compile the model
and test it out in less than one hour or so, it is pretty much a
uselessly slow research process.  Once I hone in on the "right part of
the parameter space", then I can let it run a few weeks.

If you run heatbugs, you can change the space size and the number of
bugs to get an idea.  in an xterm run the "top" program while you
experiment.


Tee Toth-Fejel wrote:
> 
> Undoubtedly, people have asked the question,
> "How many agents can swarm run on a PC?"
> The answer obviously depends on the size and complexity of agents, and of
> their interactions.  But what are the traded offs?
> 
> Has anyone done this sort of experimentation?
> 
> Thanks!
> 
> Tihamer "Tee" Toth-Fejel                    Member of Technical Staff


-- 
Paul E. Johnson                       email: address@hidden
Dept. of Political Science            http://lark.cc.ukans.edu/~pauljohn
University of Kansas                  Office: (785) 864-9086
Lawrence, Kansas 66045                FAX: (785) 864-5700

                  ==================================
   Swarm-Support is for discussion of the technical details of the day
   to day usage of Swarm.  For list administration needs (esp.
   [un]subscribing), please send a message to <address@hidden>
   with "help" in the body of the message.