大规模分布式仿真中实体分配策略研究

在大规模分布式仿真中，能否为数量众多的仿真实体合理指定其所在的宿主仿真机将直接影响系统性能。本文在文献[1]提出的实体分配评价指标以及仿真场景图划分代价表示的基础上，首先介绍了仿真场景图的递归对分，接着提出了两种实体分配算法：最大流量顶点归并法和最大可能收益顶点归并法。针对均匀分布、倾斜分布、聚集分布以及随机分布等各种实体分布类型，分别评测了各算法的配置结果和代价。评测结果表明，最大可能收益顶点归并法能得到较优结果，且执行速度快。

Entity Partitioning in Large Scale Distributed Simulations

In large scale distributed simulations, entity partitioning has great impact on system performance. Based on e-valuation indices and the cost model presented in [1], this article first describes the Recursive Bisection of Simulation Scene Map. Then two entity partitioning algorithms, Maximum Flow Node Merging and Maximum Possible Gain Node Merging, are proposed. The performances of these algorithms are analyzed and tested in several scenes with different distribution of entities, including uniform distribution, skewed distribution, clustered distribution and random distribution. The simulated results show that the algorithm of Maximum Possible Node Merging is better than its counterparts.