[Swarm-Modelling] Question about Distributed Simulations

[Andy Cleary]

Unfortunately, the work that I was doing that John mentions was a
victim of the infathomable and maddening black art of "government
funding" and did not get to a point where it could be used for these
purposes, and the difficulty of trying to use a machine like BlueGene-L
for an application person (that is: non-computer scientist) without a
package to help you is probably insurmountable unless your simulations
have some really, really nice properties (and most interesting
simulations do not have these properties). 

I'm certainly interested in hearing what a surgeon is trying to model
that requires BG/L; part of the reason for the loss of my funding was a
question over the real-world need for ABM on such machines. The
argument that there was clearly *going* to be a need for it and that to
wait until that need knocked on your doorstep before beginning that
work would mean a several year delay apparently wasn't very convincing,
though it appears to be coming to pass. Perhaps you can partner with a
CS team at one of the labs now and be that driver, though that will
mean that you'll have a delay of several years waiting for the software
to catch up, and by that time *you'll* probably lose your funding.

Cheers,
Andy, who has no bitterness towards the government funding process at all... ok, maybe a *little*. ;-)

p.s. Syncer was neat stuff. 

-- 
Dr Andrew J. Cleary
Director of Algorithms R&D
Visible Path Corporation
www.visiblepath.com

From: 
address@hidden
To: Agent-based modeling <address@hidden>
Date: Tue, 22 Aug 2006 09:51:40 -0400
Subject: Re: [Swarm-Modelling] Question about Distributed Simulations
As others will confirm, what you are asking to do is a hard computer
science problem. We developed some Java code for doing what you want with
Swarm called Syncer back in 2000. Unfortunately it was never updated with
the newer releases of Swarm and the last time we checked it out, it would
not run with Swarm any longer. I would be willing to send you the paper
describing the algorithm and the source code if you want to try and get it
to work with the current releases.

You can also check out a paper by Matthias Scheutz entitled, "Adaptive
Algorithms for the Dynamic Distribution and Parallel Execution of
Agent-Based Models". A copy can be found at:
http://www.nd.edu/~airolab/publications/scheutzschermerhorn06pardist.pdf

Finally, back in '02 I was in communication with Andy Cleary at Lawrence
Livermore who haunts these mailing lists on occasion. He was working on
distributing Swarm over a 1000+ node tightly coupled architecture. He may
be able to help.

John A. Sauter
Business Unit Leader, Emerging Markets
ph: 734.302.4682 cell: 734.904.2313 fax: 734.302.5661
NewVectors, LLC (http://www.newvectors.net)
Innovative Solutions in Program Management, Analysis and Logistics
Street: Suite 250, 3520 Green Ct, Ann Arbor, MI 48105-1579



             address@hidden
             Sent by:
            
modelling-bounces

                             To
             @swarm.org                 address@hidden
                                                                        cc

             08/22/2006 01:15                                      Subject
             AM                        [Swarm-Modelling] Question about
                                       Distributed Simulations

             Please respond to
                Agent-based
                 modeling
             <address@hidden
                   org>






Hello list,

I'm not sure which is the correct list for this question so my
apologies in advance. I am looking into the feasibility of running a
Swarm simulation on IBM's Blue Gene L, which has as its base rack a
1000 node cluster (2000 PPC 5 processors). The question is whether an
individual run could be done across the entire cluster, with groups of
agents distributed among all the nodes and communication across the
nodes. I recall threads having to with distributed simulations but
cannot remember if that was distribution for batch runs or for a single
run. If this has been extensively discussed perhaps someone could point
me towards that archived thread? Thanks in advance.

Gary An
Department of Surgery
Northwestern University
Chicago, IL
__________