摘要
多核中央处理器(central processing units,CPU)-图形处理器(graphics processing units,GPU)异构平台为提高并行Agent仿真(parallel Agent-based simulation,PABS)在单机上的运行性能提供了一个更高效的硬件基础,但在当前相关研究中,还缺乏一般性的理论方法来指导并行Agent仿真将多核CPU和GPU的计算资源充分利用起来。通过分析多核CPU-GPU异构并行架构的特点,在方法论层面上建立了并行Agent仿真在多核CPU-GPU异构平台下的多层负载分配模型,并根据基于Agent的仿真的执行结构,提出了对基于Agent的仿真的计算结构、数据结构进行重构的方法,以适应异构的硬件架构。最后对基于多核CPU-GPU的并行Agent仿真性能进行了实验分析。
The platforms equipped with multi-eore central processing units （CPU） and graphics processing units （GPU） provide more efficient hardware for improving the performance of parallel Agent-based simulations （PABS） on personal computers. In the correlative research field, there is currently lack of a general methodolo- gy that can take full advantage of the computing capability of multi-core CPUs and GPUs for PABS. By analy- zing the characteristics of the heterogeneous architecture that consists of multi-core CPUs and GPUs, a multi- layer load distributing model for PABS on multi-core CPU and GPU platforms is built. According to the para- digm of Agent-based simulations （ABS）, the methods of rebuilding the computing and data structures of ABS are presented so as to adapt for the heterogeneous architecture. Finally a group of experiments are carried out for testing the performance of PABS on multi-core CPU and GPU platforms.
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2012年第8期1716-1722,共7页
Systems Engineering and Electronics
基金
国家自然科学基金(60974073
60974074)资助课题
关键词
并行Agent仿真
多核中央处理器
图形处理器
异构平台
parallel agent-based simulation （PABS）
multi core ceritral processing units （CPU）
graphics processing unit （GPU）
heterogeneous platform
