10 likes | 166 Views
On-the-fly parallelization in Agent-Based Simulation Systems. Cole Sherer Computer Science Department The University of Georgia koleslaw@uga.edu. Computer Science Department The University of Georgia. Abstract
E N D
On-the-fly parallelization in Agent-Based Simulation Systems Cole Sherer Computer Science Department The University of Georgia koleslaw@uga.edu Computer Science Department The University of Georgia • Abstract • Agent-based simulation systems (ABS) are increasingly being used to solve a wide-array of problems in business, telecommunications, robotics, computer games, and military applications.1 ABS modelers face two challenges: First, performance is affected as their simulations become more complex and larger scale; Second development is difficult because there is no common interface to the array of platforms that support ABS work. • We seek to transform popular, intuitive, sequential ABS APIs into more efficient PDES code automatically. As a first step we are ‘parallelizing’ the popular MASON multiagent simulation kit developed at George Mason University. Other potential targets include Player/Stage and Teambots. • Related Work • MASON is a very popular ABS written in Java, but it is designed to run sequentially.2 • Teambots is developed at Georgia Tech and has been used by hundreds of researchers across the world. It is also written in Java and is designed to run sequentially.2 • SASSY is our Java-based PDES with an ABS API built on top.3 • Many compilers have been developed to tackle general auto-parallelization, but perform poorly. • Focusing on ABS systems will give us an advantage over the general approach. • Approach • Compiler-like approach using a preprocessor • Build abstract syntax tree4 • Consult lookup dictionary for conversion rules • Rewrite Java source to run in SASSY • Starting with MASON, but can swap out lookup dictionary to other systems in future (e.g., RePast, Player/Stage, and Teambots) • Milestones • Develop SASSY ABS middleware • Map core MASON classes to SASSY • Design conversion system • Fully implement the system • Experiment and test the system • Acknowledgments • George Mason University’s Evolutionary Computation Lab and Center for Social Complexity for the MASON toolkit. • Dr. Maria Hybinette for advising me in this research. • References • Logan, B. and Theodoropoulos, G. The Distributed Simulation of Multiagent Systems, Proceedings of the IEEE, vol 89, no 2, 2001. • Sean Luke et al. MASON: A Multi-Agent Simulation Environment, Transactions of the society for Modeling and Simulation International, 2005. • Maria Hybinette et al SASSY: A Design for a Scalable Agent-based Simulation System using a Distributed Discrete Event Infrastructure, Proceedings of the 2006 Winter Simulation Conference. • Abstract Syntax Tree - http://www.eclipse.org/articles/article.php?file=Article-JavaCodeManipulation_AST/index.html