2016年12月1日
A Study of Real-time and 100 Billion Agents Simulation Using the Boids Model
Artificial Life and Robotics
- ,
- ,
- ,
- ,
- 巻
- 21
- 号
- 4
- 開始ページ
- 525
- 終了ページ
- 530
- 記述言語
- 英語
- 掲載種別
- 研究論文(学術雑誌)
- DOI
- 10.1007/s10015-016-0308-3
- 出版者・発行元
- Springer Tokyo
In high-performance computing of multi-agent systems, there often exists a load imbalance that slows down the calculation. In this paper, we discuss the parallelization of the Boids model for simulating a swarm intelligence. We apply the multi-level hierarchy of parallelism to the Boids model to mitigate the load-imbalance problem. To eliminate numerical errors due to parallelization, we apply pseudo-quadruple arithmetic. The parallel performance is evaluated on three major architectures, including many-core processors on an x86-based server with GPUs, and the Earth Simulator. The parallelization can decrease the negative effects of a load imbalance to almost zero in a simulation of 50 million agents. In addition, the parallelization guarantees the reproducibility of the results in a sequential execution. The strong scaling shows the potential to complete a simulation in real-time on the Earth Simulator. In addition, the weak scaling shows the ability to calculate 100 billion agents within a reasonable amount of time.
- ID情報
-
- DOI : 10.1007/s10015-016-0308-3
- ISSN : 1614-7456
- ISSN : 1433-5298
- SCOPUS ID : 84982099105