一种基于任务区分的仿真资源调度策略研究

发布时间：2018-04-29 01:30

本文选题：巨型机 + 仿真　；参考：《计算机仿真》2013年04期

【摘要】：在如今的大规模并行计算中,一般都会使用巨型机作为仿真工具,而巨型机计算资源的有效调度能很大程度地提高仿真计算速度和资源的使用效率。首先针对当今主流巨型机的结构特点,设计了一种集中控制分布式代理部署的并行计算资源调度结构,然后根据系统提交审核的仿真计算作业的任务重要性,提出了一种根据任务区分的计算资源自适应调度策略机制和实现策略机制的主要关键算法。通过并行计算仿真的对比和分析,验证了调度策略有较好的实际应用效果。
[Abstract]:In today's large-scale parallel computing, supercomputers are generally used as simulation tools, and the efficient scheduling of supercomputer computing resources can greatly improve the speed of simulation computation and the efficiency of the use of resources. Firstly, according to the structural characteristics of the mainstream supercomputers, a parallel computing resource scheduling architecture with centralized control over the deployment of distributed agents is designed, and then the task importance of the simulation computing job submitted for verification by the system is presented. In this paper, an adaptive scheduling policy mechanism for computing resources and a key algorithm for implementing the policy mechanism based on task differentiation are proposed. Through the comparison and analysis of parallel computing simulation, it is verified that the scheduling strategy has good practical application effect.
【作者单位】：第二炮兵装备研究院第二研究所;
【基金】：第二炮兵装备研究院青年创新基金
【分类号】：TP393.01

【参考文献】