基于聚簇的大规模虚拟试验实体分配策略研究

在大规模虚拟试验中,能否为数量众多的试验实体合理指定其所在的宿主机将直接影响系统性能.针对现有实体分配算法大都以线性聚簇策略为主,不适用于大规模松耦合的试验环境,提出了一个较为系统性的细粒度实体分配理论.在网络仿真实时性理论指导下,建立了实体交互图EIG,定义了虚拟试验系统的实体粒度.提出了大规模试验中非线性聚簇的实体划分策略,并证明了细粒度下非线性聚簇要优于线性聚簇.该实体分配理论对实体聚合算法的选取和分配方案的评价都具有指导意义.

Entity Partitioning Based on Clustering in Large Scale Virtual Testing System

In large scale virtual testing system, the overall performance is up to the quality of host assignment for each entity. Given the fact that current entity partitioning algorithms use linearly clustering strategy and cannot meet the requirement of large scale virtual testing system, a systematic granularity theory of fine-grained entity partitioning is proposed. The entities interactive graph is created under the theory of network real-time simulation, and the granularity of entities is defined. The partitioning strategy of nonlinear clustering is raised. It is proven that the nonlinear clustering is better than linear one in large scale virtual testing system. This theory is applicable to choosing entity merging algorithms and evaluating partitioning plans.