相对于时间步进仿真，具有更高运行效率的并行离散事件仿真方法近年来逐渐成为仿真界的研究热点。但事件的随机性和不规则性使后者更难于对应到GPU众核的大规模并行性，为了克服这一困难，本课题提出研究一种面向并行离散事件仿真的基于CPU和GPU多线程协同调度与通信的方法。同时，为克服传统保守时间推进策略事件处理并行性不足的缺点，提出一种乐观／保守并存的利用运行时信息扩展可安全执行事件时间界限的算法。进而从改进仿真内存布局和提高系统计算／通信比出发，提出一种减小GPU线程核全局内存访问延迟的算法。从而全面提高CPU／GPU异构资源利用率，为解决大规模并行离散事件仿真系统运行效率低的问题打下坚实基础。

Compared to time-stepped simulation, Parallel Discrete Event Simulation (PDES) is more efficient and has attracted more research interests nowadays. However, it is more difficult for PDES to be applied with GPU than time-stepped simulation because PDES has many stochastic events. To overcome this shortcoming, this project proposes to study a method to coordinate CPU and GPU via multi-thread scheduling and communication. Moreover, to promote parallelism of traditional conservative time advancement strategy, an algorithm maintaining both pessimistic and conservative characteristics is proposed using running information to enlarge limit of events which can be safely executed. And to enhance GPU memory layout and computation-to-communication ratios, an approach is to be studied to avoid latency caused by memory bandwidth saturation. All the above works can gain a higher utilization of heterogeneous resources of CPU and GPU, and provide a solid base for resolving the problem of low running efficiency of large scale PDES systems.