并行Agent仿真研究综述

Review of Parallel Agent-based Simulation

作为一种研究复杂系统的有效途径,基于Agent的建模仿真方法得到了广泛应用,但随着应用规模和复杂度的增加,仿真运行速度成为制约其应用的一个重要因素。并行Agent仿真通过将Agent模型分配到多个处理单元上同时运行来减少仿真运行时间,是提高仿真运行速度的一个直接手段。根据基于Agent的建模仿真的基本思想,分析了并行Agent仿真区别于传统的并行离散事件仿真的新特点,着重从时间同步协议、负载均衡、通信优化三方面阐述了并行Agent仿真的研究现状,并对基于通用目的图形处理器的并行Agent仿真进行了讨论。最后对并行Agent仿真未来的研究方向进行了展望。

As an effective approach to study complex systems, the method of Agent-based Modeling and Simulation （ABMS） is used widely in many research fields, but as the scale and complexity of its applications enlarge, the execution speed becomes an obstacle to implement ABMS. Parallel Agent-based Simulation （PABS） aims at reducing the execution time through executing concurrently the Agent models distributed on different process units, which is a direct approach to improve the execution speed. According to the principles of ABMS, the new characteristics of PABS that are different from Parallel Discrete Event Simulation were proposed. The overview about PABS was elaborated from the aspects of time synchronization algorithms, load balancing and communication optimizing. PABS based on general purpose graphics processing units was discussed. At last, some future research directions about PABS were proposed.