DS-RT '12 Proceedings of the 2012 IEEE/ACM 16th International Symposium on Distributed Simulation and Real Time Applications
| Hi-index | 0.00 |
This paper describes highly scalable X10-based agent simulation platform called XAXIS. XAXIS is designed to handle millions or billions of agents on recent highly distributed and parallel computing environments with more than hundreds of CPU cores. To make the runtime scalable on such environments, we need to redesign and implement the simulation middleware. In this paper, we propose the software design, implementation on X10, one of the state-of-the-art PGAS language, and then application to large-scale traffic simulation. By using 192 CPU cores in distributed memory computing environment, the performance scalability is achieved with a traffic simulation. We also propose performance optimization approach that accelerates the simulation time by adaptively changing the simulation synchronization frequency while maintaining the simulation result accuracy. With the Tokyo road network data, from 3 to 4 fold speeds up are obtained with our proposed performance optimization.