[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: simulating large models
From: |
Jason Alexander |
Subject: |
Re: simulating large models |
Date: |
Mon, 10 Jun 2002 15:04:17 +0100 |
Does anyone have experience with simulating large models with swarm?
Is swarm prepared to distributed simulations across several
processors/machines?
We're thinking of large market models which may include hundreds (or
even
thousands) of customers (agents) and we foresee scalability problems.
A model including hundreds or thousands of agents will not, in
principle, present a problem unless your agents are particularly complex
or resource hungry. Swarm easily handles lattice models where each
slot in a grid is occupied by an agent and the lattice has dimensions
200x200 or 300x300 (so you have 40000 or 90000 agents existing at once).
What does become more important with large models are the algorithms you
use for manipulating the population. Inefficient algorithms which don't
seem to slow things down in small models can easily become a bottleneck
in large models. If the agents are of heterogenous types, using the
standard Swarm messaging methods can also incur a performance hit,
unless things have changed recently. One of my lattice models in which
the agents were heterogenous slowed down by an order of magnitude
because of some internal changes to Swarm's messaging methods required
to support the Java interface. However, I was able to avoid this
performance hit by using an alternative method of passing messages to
the agents.
If I remember correctly, Swarm doesn't support distributed models in a
transparent way. (That is, you write a model and Swarm distributes the
computational load across several machines.)
However, I'm always a bit curious as to why one might need to do this,
aside from the coolness factor. You can get a Linux box with 2 or 4
high-speed processors and a gig of ram (or more) for reasonable amounts
nowadays. With a box like that I would think one would be able to do an
awful lot, especially if one wrote the model efficiently. (I.e., instead
of having 24 separate boolean variables, use a single int with bit
arithmetic.)
Cheers,
Jason
--
J. McKenzie Alexander
Department of Philosophy, Logic and Scientific Method
London School of Economics and Political Science
Houghton Street, London WC2A 2AE
==================================
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.