Agent-based model: Difference between revisions

Revision as of 11:00, 13 March 2024

Revision as of 22:27, 23 March 2024

Line 135:

==Implementation==

Many [[Comparison of agent-based modeling software|ABM frameworks]] are designed for serial [[Von Neumann architecture|von-Neumann computer architectures]], limiting the speed and scalability of implemented models. Since emergent behavior in large-scale ABMs is dependent of population size,<ref name="Lysenko 2008 MegaScale">{{cite journal |last1=Lysenko |first1=Mikola |last2=D'Souza |first2=Roshan M. |title=A Framework for Megascale Agent Based Model Simulations on Graphics Processing Units |journal=Journal of Artificial Societies and Social Simulation |date=2008 |volume=11 |issue=4 |pages=10 |url=http://jasss.soc.surrey.ac.uk/11/4/10.html |access-date=16 April 2019 |issn=1460-7425 |archive-date=April 26, 2019 |archive-url=https://web.archive.org/web/20190426210422/http://jasss.soc.surrey.ac.uk/11/4/10.html |url-status=live }}</ref> scalability restrictions may hinder model validation.<ref>{{cite journal |last1=Gulyás |first1=László |last2=Szemes |first2=Gábor |last3=Kampis |first3=George |last4=de Back |first4=Walter |title=A Modeler-Friendly API for ABM Partitioning |journal=Proceedings of the ASME 2009 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference IDETC/CIE 2009 |date=2009 |volume=2 |pages=219–226 |url=https://proceedings.asmedigitalcollection.asme.org/proceeding.aspx?articleid=1649189 |___location=San Diego, California, USAUS |access-date=April 16, 2019 |archive-date=April 16, 2019 |archive-url=https://web.archive.org/web/20190416181927/https://proceedings.asmedigitalcollection.asme.org/proceeding.aspx%3Farticleid%3D1649189 |url-status=live }}</ref> Such limitations have mainly been addressed using [[distributed computing]], with frameworks such as Repast HPC<ref>{{cite journal |last1=Collier |first1=N. |last2=North |first2=M. |title=Parallel agent-based simulation with Repast for High Performance Computing |journal=Simulation |volume=89 |issue=10 |pages=1215–1235 |doi=10.1177/0037549712462620 |year=2013 |s2cid=29255621 }}</ref> specifically dedicated to these type of implementations. While such approaches map well to [[Computer cluster|cluster]] and [[Supercomputer architecture|supercomputer]] architectures, issues related to communication and synchronization,<ref>{{cite book |last1=Fujimoto |first1=R. |title=2015 Winter Simulation Conference (WSC) |chapter=Parallel and distributed simulation |date=2015 |pages=45–59 |doi=10.1109/WSC.2015.7408152 |___location=Huntington Beach, CA, USAUS |isbn=978-1-4673-9743-8 |s2cid=264924790 |chapter-url=http://www.lib.ncsu.edu/resolver/1840.4/5268 |access-date=September 6, 2020 |archive-date=February 4, 2023 |archive-url=https://web.archive.org/web/20230204160704/https://repository.lib.ncsu.edu/handle/1840.4/5268 |url-status=live }}</ref><ref>{{cite journal |last1=Shook |first1=E. |last2=Wang |first2=S. |last3=Tang |first3=W. |title=A communication-aware framework for parallel spatially explicit agent-based models |journal=International Journal of Geographical Information Science |date=2013 |volume=27 |issue=11 |pages=2160–2181 |doi=10.1080/13658816.2013.771740 |publisher=Taylor & Francis|bibcode=2013IJGIS..27.2160S |s2cid=41702653 }}</ref> as well as deployment complexity,<ref>{{cite journal |last1=Jonas |first1=E. |last2=Pu |first2=Q. |last3=Venkataraman |first3=S. |last4=Stoica |first4=I. |last5=Recht |first5=B. |title=Occupy the Cloud: Distributed Computing for the 99% |journal=Proceedings of the 2017 Symposium on Cloud Computing (SoCC '17) |date=2017 |pages=445–451 |doi=10.1145/3127479.3128601 |arxiv=1702.04024 |bibcode=2017arXiv170204024J |publisher=ACM |___location=Santa Clara, CA, USAUS|s2cid=854354 }}</ref> remain potential obstacles for their widespread adoption.

A recent development is the use of data-parallel algorithms on Graphics Processing Units [[GPU]]s for ABM simulation.<ref name="Lysenko 2008 MegaScale" /><ref>{{cite web |author=Isaac Rudomin |url=https://sites.google.com/site/rudominisaac/shader-agents |title=Large Crowds in the GPU |year=2006 |publisher=[[Monterrey Institute of Technology and Higher Education]] |display-authors=etal |url-status=dead |archive-url=https://web.archive.org/web/20140111054342/https://sites.google.com/site/rudominisaac/shader-agents |archive-date=January 11, 2014 }}</ref><ref>{{cite journal |first1=Paul |last1=Richmond |first2=Daniela M. |last2=Romano |url=http://www.dcs.shef.ac.uk/~daniela/Paul_abgpu_IWSV_2008.pdf |archive-url=https://wayback.archive-it.org/all/20090115220835/http://www.dcs.shef.ac.uk/~daniela/Paul_abgpu_IWSV_2008.pdf |url-status=dead |archive-date=January 15, 2009 |title=Agent Based GPU, a Real-time 3D Simulation and Interactive Visualisation Framework for Massive Agent Based Modelling on the GPU |journal=Proceedings International Workshop on Super Visualisation (IWSV08) |year=2008 |access-date=April 27, 2012 |df=mdy-all }}</ref> The extreme memory bandwidth combined with the sheer number crunching power of multi-processor GPUs has enabled simulation of millions of agents at tens of frames per second.