【作者】 刘步权；

【导师】 王怀民；

【作者基本信息】 国防科学技术大学 ， 计算机科学与技术， 2004， 博士

【摘要】 高层体系结构HLA(High Level Architecture)是美国国防部为了解决在各个领域开发出来的多种模型、仿真应用和C4I系统的互操作和可重用问题而提出的一种分布式仿真规范。2000年9月，HLA被正式接纳为IEEE 1516标准。HLA中的时间管理服务是在综合了以往各类仿真中的时间管理技术的基础上提出来的，对于解决分布式仿真中消息的时序问题和保证仿真的正确执行具有十分重要的意义。 时间管理服务是HLA／RTI中的核心技术，时间管理服务的基本原理和关键算法的研究是当前分布式仿真领域公认的热点和难点问题。特别是随着分布式仿真规模的不断扩大，时间管理服务的实际应用也受到了空前的挑战。 本文重点针对大规模分布式仿真环境下高效的时间管理服务展开研究，其重要贡献在于从理论和实现等方面系统地研究了IEEE 1516标准中时间管理服务的基本原理、关键算法和在基于CORBA(Common Object Request Broker Architecture)的层次式RTI(Runtime Infrastructure)软件StarLink+中的具体实现机制。 最大可用逻辑时间GALT(Greatest Available Logical Time)在HLA 1.3中被称之为LBTS(Lower Bound Time Stamp)，是实现时间管理服务的关键。本文针对GALT算法中可能存在的死锁问题，系统地分析了GALT的计算要素，剖析了相关算法的局限性，揭示了死锁产生时系统所特有的规律，提出了两种无死锁的GALT算法，即身高测量法1和身高测量法2，并从理论上对算法的正确性进行了证明。身高测量法为正确实现遵循IEEE 1516标准的时间管理服务提供了可靠保证。 HLA中的乐观推进机制来源于并行离散事件仿真PDES(Parallel Discrete Event Simulation)，但又有所不同。本文针对乐观推进机制中回卷管理复杂低效的问题，在深入研究PDES和HLA两种乐观推进机制的基础上，对两种不同的推进机制进行了分析和比较，描述了它们之间的重要差异，给出了仿真应用程序的开发方法，提出了不需要RTI做任何保存操作的“零保存”机制。“零保存”机制大大地简化了RTI中时间管理服务的实现。 层次式体系结构的RTI能够有效地降低时间管理服务的计算开销，为大规模仿真应用提供有效的支持。StarLink+的层次式体系结构包括中心RTI服务器、局部RTI服务器和盟员三个层次。本文针对大规模分布式仿真中时间管理服务可能带来的巨大开销，基于层次式RTI的体系结构，提出并实现了遵循IEEE 1516标准的“集中式”和“分布式”两套时间管理服务方案，并集成在同一个软件包中可供用户通过配置文件进行选择。本文详细地描述了两种时间管理实现方案中每个重要服务的实现算法。 在集中式时间管理实现中，提出了基于“最小二元法”的双层RTI并行推进机制，盟员的时间推进由中心RTI服务器和局部RTI服务器协同完成，这种决策机制可有效地提高时间管理服务的决策效率。盟员在请求时间推进时，首先由所属的局国防科学技术大学研究生院学位论文部RTI服务器快速决策;若局部RTI服务器无法推进则由中心RTI服务器进行全面协调;中心Rl，1服务器拥有计算GALT的全部信息，中心Rl，1服务器通过“最小二元法”选取两个最小的盟员身高发送到局部RTI服务器，则局部RTI服务器即可计算出所有本地盟员的GALT。 在分布式时间管理实现中，提出了可由局部RTI服务器独立计算GALT的分布并行推进机制。盟员的时间推进由局部Rl，1服务器独立完成，中心Rl，I服务器并不参与时间管理服务的决策功能，每个局部RTI服务器只相当于其它局部RTI服务器的一个普通盟员，这种决策机制可有效地减少GALT计算时各个局部RTI服务器之间的通信开销，为实现使用HLA时间管理服务的大规模分布式仿真应用提供有效保证。 论文通过各种试验对StarLink十的时间管理服务进行了系统的测试，结果表明:StarLink+时间管理服务的功能完全，性能优良，支持大规模仿真应用，具有跨路由和广域网运行能力。特别重要的是，starLink十的时间管理服务已经在由1 800个盟员构成的仿真系统中成功地得到了检验。 本文提出的关键算法和实现机制对于其它RTI软件的研究同样具有重要的参考价值。关键词:高层体系结构，运行支撑平台，CORBA，时间管理服务，最大可用逻辑时 间，身高测量法，零保存第11贡更多还原

【Abstract】 High Level Architecture (HLA) is a distributed modeling and simulation standard put forward by American Department of Defense, which is used to facilitate interoperability and reuse of various models and simulations among themselves and with C4I-systems. HLA was adopted as the IEEE 1516 standard in September 2000. Time Management services in HLA is brought forward on the basis of all kinds of former time management technologies. The main goal of time management services is used to resolve time stamp order of messages in a distributed simulation, and ensures that the simulation can be executed correctly.Time management services is the critical technology in HLA/RTI, while the research on the principles and key algorithms in time management services is a well-known important and intractable problem in present modeling and simulation area. Especially with the frequent rise of simulation scales, the actual application of time management services is also meeting an unprecedented challenge.The objective of the dissertation is to study the efficient time management services for large-scale distributed simulations. The major contributions of this thesis focuses on the principles and key algorithms in time management services according to the IEEE 1516 standard, and the implementation mechanisms in a hierarchical RTI (Runtime Infrastructure) software named StarLink+ based on CORBA (Common Object Request Broker Architecture).It is most important for GALT (Greatest Available Logical Time) algorithm to implement time management services which is also called LBTS (Lower Bound Time Stamp) in HLA 1.3. In order to resolve the deadlock which may exist in a GALT algorithm, this thesis systematically analyzes the factors of GALT computation, anatomizes a few of algorithms’ shortages, reveals some important conclusions about the deadlock, and Two GALT algorithms without deadlock are put forward, which are called stature-measuring I and stature-measuring II. The correctness of the two algorithms are also proved, and they can exactly provide reliable technology support to develop time management services according to IEEE 1516.The optimistic advancing mechanism in HLA originally comes from PDES (Parallel Discrete Event Simulation), but they are not always the same. In order to resolve the complicated implementation of rollback mechanism with low efficiency, this thesis analyzes and compares the two different optimistic advancing mechanisms in PDES and HLA, reveals some important differences between them, explains the programming method for a simulation application and the Zero-Saving mechanism without saving any operations in a RTI is proposed. The Zero-Saving mechanism can simplify the development of time management services in a RTI greatly.A hierarchical RTI can effectively decrease the computation expenditure in timemanagement services, and support the large-scale simulations. The hierarchical architecture in StarLink+ is composed of three levels that are the central RTI server (CRTI), the local RTI servers (LRTI), and the joined federates. In order to resolve the great expenditure in time management services for large-scale simulations, this thesis proposes and implements two schemes in time management services according to IEEE 1516 standard based on the hierarchical RTI architecture. One is the central scheme and the other is the distributed scheme. The dissertation also describes the algorithms of all important time management services in the two schemes in details.The double-level parallel advancing technology based on the two least elements is put forward in the central scheme. Time advancement is controlled by the central RTI server and all local RTI servers, and the technology can greatly enhance the decision-making efficiency. When a joined federate requests to advance time, its local RTI server firstly makes rapid decision and decides whether the joined federate can be granted to advance logical time. If the local RTI server suspends the request, the central RTI server must dispose the request globally. As it has all informa更多还原