swarm-modeling
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Distributed Simulations with Swarm


From: Marcus G. Daniels
Subject: Re: Distributed Simulations with Swarm
Date: 12 Oct 2000 09:52:05 -0700
User-agent: Gnus/5.070084 (Pterodactyl Gnus v0.84) Emacs/20.4

>>>>> "MG" == gaston  <address@hidden> writes:

MG> Does anyone have any information on distributed
MG> agent-based simulation using Swarm? 

Swarm is a kind of discrete event simulator that merges agents'
schedules into a single time sequence.  In this time sequence all actions
that co-occur are tagged as such.  One thing we would like to do with
Swarm would be to have different ways to distribute those co-occuring
actions to different CPUs or separate computers, e.g. via threads and by MPI.

You might also be interested in a package called Drone that can be
used to distribute whole model runs to different computers.  Drone has
features for parameter sweeks, etc.  Another option is to use a
clustering operating system like MOSIX or the new Beowulf systems.  In
that model, you just have a shell script that starts spawning models
with different initial conditions, and the operating system takes care
to distribute and balance the load to different computers.  (Like SMP,
but cross-box...)

Finally, Swarm has a fairly complete object serialization capability,
so you can save an object on one machine and reload it on another.
There are two backends to this, one that is stream-oriented and
another that is random access, but can run on MPI.

MG> For instance, has there been any work on developing an HLA wrapper
MG> for swarm (or even objc)?

The current source snapshots have support for COM, a component
technology that has a fair amount in common with CORBA, which HLA
uses.  For example, the Swarm interfaces are represented as IDL.  It
wouldn't be terribly difficult (but not trivial) to add support for
CORBA in Swarm.  (CORBA also would be nice to have since it is the
foundation of the GNOME desktop.)



                  ==================================
   Swarm-Modelling is for discussion of Simulation and Modelling techniques
   esp. using Swarm.  For list administration needs (esp. [un]subscribing),
   please send a message to <address@hidden> with "help" in the
   body of the message.
                  ==================================


reply via email to

[Prev in Thread] Current Thread [Next in Thread]