In large-scale distributed simulation, thousands of objects keep moving and interacting in a virtual environment, which produces a mass of messages. High level architecture (HLA) is the prevailing stan- dard for modeling and simulation. It specifies two publish-subscribe mechanisms for message filtering: class-based and value-based. However, the two mechanisms can only judge whether a message is relevant to a subscriber or not. Lacking of the ability to evaluate the relevance, all relevant messages are delivered with the same priority even when congestion occurs. It significantly limits the scalability and performance of distributed simulation. Aiming to solve the relevance evaluation problem, speed up message filtering, and filter more unnecessary messages, a new relevance evaluation mechanism Layer of Interest (LoI) was proposed by this paper. LoI defines a relevance classifier based on the impact of spatial distance on receiving attributes and attribute values. An adaptive publish-subscribe scheme was built on the basis of LoI. This scheme can abandon most irrelevant messages directly. Run-time infrastructure (RTI) can also apply congestion control by reducing the frequency of sending or receiving object messages based on each objects’ LoI. The experiment results verify the efficiency of message filtering and RTI congestion control.
ZHOU Zhong1,2 & WU Wei1,2 1 State Key Laboratory of Virtual Reality Technology and Systems, Beihang University, Beijing 100191, China
AoI(area of interest)兴趣过滤根据虚拟环境中数据的相关性进行过滤,它能有效降低网络通信量,提高系统可扩展性。在多人在线游戏中,随着虚拟环境中实体规模的增大,服务器进行兴趣过滤的计算开销会显著增大,影响游戏的可扩展性。针对近年来成本大幅度降低的多核CPU(central processing unit)架构,使用OpenMP编程接口将兴趣过滤计算的负载分配到多个CPU处理单元上进行并行计算,实验结果表明该并行算法在多核CPU架构上可获得较大的性能提升。
In distributed simulation,there is no uniform physical clock. And delay cannot be estimated because of jitter. So simulation time synchronization is essential for the event consistency among nodes. This paper investigates time synchronization algorithms over large-scale distributed nodes,analyzes LBTS(lower bound time stamp) computation model described in IEEE HLA standard,and then presents a grouped LBTS model. In fact,there is a default premise for existing algorithms that control packets must be delivered via reliable transportation. Although,a theorem of time synchronization message's reliability is proposed,which proves that only those control messages that constrain time advance need reliability. It breaks out the default premise for reliability. Then multicast is introduced into the transmission of control messages,and algorithm MCTS(multi-node coordination time synchronization) is proposed based on multicast. MCTS not only promotes the time advance efficiency,but also reduces the occupied network bandwidth. Experiment results demonstrate that the algorithm is better than others in both time advance speed and occupied network bandwidth. Its time advance speed is about 50 times per second when there are 1000 nodes,approximately equal to that of similar systems when there are 100 nodes.
ZHAO QinPing1,2,ZHOU Zhong1,2 & Lü Fang1,2 1 State Key Laboratory of Virtual Reality Technology and Systems,Beihang University,Beijing 100083,China
分布式仿真中难于实现统一的全局物理时钟,且节点之间的传输存在抖动,因此为了保证各仿真节点的事件一致性,需要进行仿真时间同步.论文对大规模分布节点的时间同步算法开展研究,对HLA标准中定义的LBTS算法进行性质分析,给出分组管理的LBTS计算模型.针对现有时间同步算法将可靠控制报文作为默认前提的问题,给出了时间控制报文可靠性定理,证明了只有制约系统推进的一部分报文需要可靠,该默认前提并非必要.在此基础上,设计了MCTS(multi-node coordination time synchronization)算法,根据时间控制报文可靠性定理引入IP组播来处理控制报文的传输,提高了节点的处理效率,大大降低了控制报文的带宽开销.实验结果表明,该算法在节点时间推进速度及网络带宽占用量方面优于同类算法,1000个节点的时间同步推进速度可达到50次/s,相当于同类系统中100个节点时的推进速度.
大规模分布式仿真中大量对象在同一个虚拟环境中运动与交互,会持续产生巨大数量的报文.HLA是建模与仿真的国际标准,它规定了基于类与基于值的发布/订购机制,但这两种机制只确定了报文中的数据与接收方的相关与否.由于缺乏报文相关性的评价,接收方需要对所有的相关报文按照相同的重要性进行处理,限制了仿真规模与效率.针对数据的相关性缺乏评价问题,文中提出可根据空间位置对属性和属性值接收的影响对数据的相关性分类,从而进行评价,提出了兴趣层次LoI(layer of interest).在LoI的基础上建立了一种自适应的发布/订购机制,它可以直接识别并抛弃大多数的不相关报文.RTI基于LoI进行报文更新和反映率的控制可以在保障重要数据的基础上进行RTI拥塞控制.BH RTI中完整实现了LoI,简介了实现机制,最后进行了报文过滤效率和RTI拥塞控制实验.
