【作者】 姚益平；

【导师】 卢锡城；

【作者基本信息】 国防科学技术大学 ， 计算机科学与技术， 2003， 博士

【摘要】 高层体系结构HLA(High Level Architecture)已于2000年9月被定为国际分布仿真通用标准IEEE1516。它把计算机及相关领域的先进技术如面向对象技术等引入分布仿真中，促进了分布式交互仿真技术的发展。随着分布式仿真应用规模的不断扩大，目前的分布式仿真运行支撑平台的性能己越来越不能满足应用的需求。为支持大规模分布式交互仿真，高性能运行支撑平台RTI(RunTime Infrastructure)实现技术正成为当前分布式交互仿真领域的研究热点，其核心问题是减少网络信息流量，提高RTI运行效率，进而解决RTI的性能瓶颈。为此，从RTI体系结构至RTI核心算法包括时间管理算法、数据分发管理区域匹配算法、对象实例发现算法等都必须加以更深入的研究。 论文主要工作是结合当前分布式交互仿真技术及相关计算机技术的发展，深入研究符合IEEE 1516标准的高性能分布式交互仿真运行支撑平台关键技术。 论文首先简要介绍了HLA／RTI的定义及组成，分析了现有几种RTI实现结构及存在的问题，针对广域网环境下大规模分布式交互仿真的需要，提出了基于CORBA及内部互操作协议的层次式RTI服务器的体系结构及设计方法。该结构具有如下特点：可实现不同软硬件平台之间的互操作；系统内部不同RTI之间通信直接、效率高；部署灵活，可扩缩性强；设置中心RTI服务器负责全局操作，为系统数据的一致性提供了保障；各局部RTI服务器可以并发执行，因而可显著提高广域网环境下大规模仿真的效率。 时间管理服务是HLA与以前分布式仿真标准最主要区别内容之一，其性能直接影响到仿真的效率和正确性。时间管理服务实现的关键是最大可用逻辑时间GALT的计算。论文针对Frederick等人提出的GALT算法存在的可能死锁问题，提出了计算GALT的递归式算法——R-GALT算法和递归式推进检测方法，并给予了无死锁证明。该算法不但解决了时间管理实现中可能出现的联盟时间推进死锁问题，而且能够提高盟员时间推进的效率。 数据分发管理实现基于值的过滤，可进一步减少大规模仿真中盟员接收冗余数据的可能性和网络中的数据流量。高效、精确的区域匹配一直是数据分发管理追求的目标。论文在分析几种区域匹配算法优缺点的基础上，提出了一种基于移动相交信息的快速区域匹配算法——RMIB算法。该算法特点是：将范围的上下界分别用一个节点表示，使用索引有序表来存放每一维上的范围节点：利用区域移动前后的信息，将匹配限定在移动范围之内，从而有效地减少了需要进行匹配的候选范围，进而减少了匹配计算量，实现了高效、精确的匹配。该算法尤其适合区域比较多的大规模分布式仿真的需要。国防科学技术大学研究生院学位论文 对象实例发现是通过对象实例交换消息的前提。论文分析了对象管理中对象实例发现的有关概念及引发实例发现的时机，在此基础上给出了带区域登记对象实例和带区域订购对象类属性时的对象实例发现算法—ROWUP算法和SOWDOWN算法，并证明了其正确性。这两个算法分别使用启发信息和剪枝条件搜索公布/订购树，提高了发现对象的搜索效率。 为了避免死锁，提高效率，最大限度地减少盟员及RTI在调用服务及进行回调时的等待时间，论文针对基于CORBA的RTI的具体实现，提出了盟员调用RTI及RTI回调盟员的多线程双缓冲并行处理技术，从而有效地提高了盟员及RTI的性能。 本文取得的研究成果已直接用于我们基于CORBA设计的符合IEEE 1516标准的RTI软件StarLink中。测试表明，StarLink各主要性能达到了国际上同类软件的水平，个别性能(如时间管理)优于国际上同类软件。关键词:高层体系结构、运行支撑平台、CORBA、最大可用逻辑时间、区域匹配、对象实 例发现、多线程、互操作第11页更多还原

【Abstract】 High Level Architecture (HLA) was approved for IEEE standard 1516 on September 2000. It introduces many advanced technologies into distributed simulation and makes the distributed interactive simulation technology develop rapidly. With the scale of distributed interactive simulation system getting larger and larger , the performance of current RunTime Infrastructure(RTI) cannot satisfy the increasing need of the applications, and the implementation technology of the high performance RTI has become a focus in distributed simulation domain. The key problem of it is to decrease the volume of information in network, and to enhance the efficiency of RTI, thereby solve the performance bottleneck of RTI. Nowadays, people are doing much more researches on the architecture and core algorithms such as Time Management, region matching, and object instance discovery of RTI.Based on the newest development of distributed interactive simulation technology and computer technology, the paper focuses on the research of key technologies of high performance RTI conformable to IEEE1516.Firstly, the definitions and components of HLA/RTI are introduced briefly, the implementation architectures and their existing problems of the current RTIs are analyzed; then aiming at the requirements of large-scale distributed interactive simulation based on WAN (Wide Area Network), the design of hierarchical RTI servers based on CORBA and interoperability protocol is put forward. The hierarchical RTI servers can get interoperability between various software and hardware platforms. Various RTIs in the system can communicate directly and efficiently. The deployment of the system is flexible. Because there is a central RTI responsible for the global operation, the global data consistency of the system can be ensured. Since all RTI servers in the system can execute in parallel, the efficiency of large-scale simulation which uses the RTI on WAN can be greatly enhanced.Time management service is one of the main features which distinguish HLA from previous distributed interactive simulation standards. Its implementation method can affect the efficiency and validity of simulation directly. The key problem of the implementation of time management service is the computation of Greatest Available Logical Time (GALT). Aiming at the possible deadlock problem of the current GALT algorithm, an optimized algorithm called recursive GALT (R-GALT) algorithm along with the idea of recursive advancing check is proposed. A formalized verification is also presented. Comparing with the original algorithm, the new algorithm can not only enhance the efficiency, but also avoid deadlock.To decrease the redundant data received by federate and the volume of information in network, data distribution management implements filtration based on value. To implement precise region matching with great efficiency is the goal of the implementation of data distribution management. After analyzing the advantages and disadvantages of several algorithms of region matching, a more effective matching algorithm called RMIB (Region Moving Information Based) algorithm is proposed. Its main features are as follows: representingthe upper bound and lower bound of a range with two nodes and storing all the nodes of the ranges of a dimension into an indexed ordered table, using the information before and after moving a region and limiting the matching computation only in the area of moving. Thereby it greatly decreases the candidate ranges that need to do matching computing, and therefore decreases the matching complexity, and obtains precise matching with high performance. This algorithm is extremely suitable for the large-scale distributed simulation that has a large number of regions.Object instance discovery is the precondition of data exchange via object instance. Related concepts of object instance discovery are analyzed and the key problems of object instance discovery are studied, then two efficient discovery algorithms named ROWUP Algorithm and SOWDOWN Algorithm are presented: one is used when reg更多还原