EpiSims (http://ndssl.vbi.vt.edu/episims.html) is a distributed discrete event ABM that runs on clusters (and soon on clusters of clusters on the TeraGrid: http://www.isdsjournal.org/article/view/1947). It is entirely possible to get slightly different results from to subsequent EpiSims runs using the same input data sets. As MGD points out in a previous message, parallelization can randomize the order of execution of events that are scheduled to run at the same future point in time. We have studied the "noise" produced by randomized execution order of same-time events in EpiSims, and found that they produce variations of results that are on the order of < 1% for most cases.
--Doug -- Doug Roberts, RTI International [EMAIL PROTECTED] [EMAIL PROTECTED] 505-455-7333 - Office 505-670-8195 - Cell On 10/11/07, David Mirly <[EMAIL PROTECTED]> wrote: > > I am currently in an Agent Based Simulation class and I am going to > do a report comparing and contrasting ABS > in parallel (distributed, etc.) environments vs. running a simulation > in a purely sequential environment. > > It seems obvious to me that you could get very different results from > one computational architecture vs. another. > > Does anyone have any experience with truly parallel systems in this > regard they would like to share? > > Thanks! > > > > On Oct 10, 2007, at 7:43 PM, Owen Densmore wrote: > > > "Super computing" is facing an interesting challenge with the advent > > of multi-core, multi-memory, blade/cluster/grid systems. > > > > The issue is the architecture one uses for powerful architectures. > > It's very difficult to have a generalized system that works well over > > a number of application architectures. And the choices are becoming > > larger by the minute. The newer "blade" systems offer both multi- > > processor and shared memory systems. They can be configured as > > clusters or as a sorta many processor system looking like a single > > memory system .. far easier to program. Grid systems are popular, > > and figuring out how to adapt to the latest hardware advances. > > > > My guess is any realistic solution will be hybrid, combining the > > features of all these large scale architectures. > > > > Here's the gotcha: how does it impact the programming language used? > > One wants an "agile" multi-processor, multi-memory architecture that > > can be reconfigured for advances in hardware and software. Thus far, > > there's no silver bullet. > > > > -- Owen [EMAIL PROTECTED] > > Beer is proof that God loves us, and wants us to be happy. > > > > On Oct 10, 2007, at 9:48 AM, Tom Johnson wrote: > > > >> FYI. Following on a brief discussion Tuesday at the data mining > >> session.... > >> > >> Google & IBM giving students a distributed systems lab using Hadoop > >> > >> Posted: 09 Oct 2007 04:07 PM CDT > >> > >> By Jesse Robbins > >> > >> Google & IBM have partnered to give university students hands-on > >> experience developing software for large-scale distributed systems. > >> This initiative focuses on parallel processing for large data sets > >> using Hadoop, an open source implementation of Google's MapReduce. > >> (See Tim's earlier post about Yahoo & Hadoop ) > >> > >> "The goal of this initiative is to improve computer science > >> students' knowledge of highly parallel computing practices to > >> better address the emerging paradigm of large-scale distributed > >> computing. IBM and Google are teaming up to provide hardware, > >> software and services to augment university curricula and expand > >> research horizons. With their combined resources, the companies > >> hope to lower the financial and logistical barriers for the > >> academic community to explore this emerging model of computing." > >> The project currently includes the University of Washington, > >> Carnegie-Mellon University, MIT, Stanford, UC Berkeley and the > >> University of Maryland. Students in participating classes will have > >> access to a dedicated cluster of "several hundred computers" > >> running Linux under XEN virtualization. The project is expected to > >> expand to thousands of processors and eventually be open to > >> researchers and students at other institutions. > >> > >> As part of this effort, Google and the University of Washington > >> have released a Creative Commons licensed curriculum to help teach > >> distributed systems concepts and techniques. IBM is also providing > >> Hadoop plug-ins for Eclipse. > >> > >> Note: You can also build similar systems using Hadoop with Amazon > >> EC2 . Tom White recently posted an excellent guide and Powerset has > >> been using this in production for quite some time. > >> > >> > >> --tj > >> -- > >> ========================================== > >> J. T. Johnson > >> Institute for Analytic Journalism -- Santa Fe, NM USA > >> www.analyticjournalism.com > >> 505.577.6482(c) 505.473.9646(h) > >> http://www.jtjohnson.com [EMAIL PROTECTED] > >> > >> "You never change things by fighting the existing reality. > >> To change something, build a new model that makes the > >> existing model obsolete." > >> -- Buckminster > >> Fuller > >> ========================================== > >> ============================================================ > >> FRIAM Applied Complexity Group listserv > >> Meets Fridays 9a-11:30 at cafe at St. John's College > >> lectures, archives, unsubscribe, maps at http://www.friam.org > > > > > > ============================================================ > > FRIAM Applied Complexity Group listserv > > Meets Fridays 9a-11:30 at cafe at St. John's College > > lectures, archives, unsubscribe, maps at http://www.friam.org > > > ============================================================ > FRIAM Applied Complexity Group listserv > Meets Fridays 9a-11:30 at cafe at St. John's College > lectures, archives, unsubscribe, maps at http://www.friam.org >
============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org
