【作者】 洪志国；

【导师】 范植华；

【作者基本信息】 中国科学院研究生院（软件研究所） ， 计算机应用技术， 2005， 博士

【摘要】 本文首先概述了国内外卫星网络及其建模的研究现状,综合分析了卫星网络的建模方法,并比较了网络性能分析的常用方法。由于卫星昂贵等特点使得卫星网络的物理实验方式可行性不高,所以最佳的选择是把数学建模方法与仿真分析方法相结合,实现优势互补。本文主要把建模分析和仿真分析相结合来分析和研究卫星网络的性能,为卫星网络的组建提供重要的指导和参考。 国内外对卫星网络的性能评价工具目前多集中在排队论和马尔可夫过程。然而,排队模型不能刻画卫星网络中信息业务的并行性、异步性和分布性的特点。SPN(Stochastic Petri Nets,随机Petri网)的描述能力强于排队网络,且克服了用马尔可夫过程分析时难以在相应的随机过程层次上建模的缺点。因此,本文采用SPN对卫星网络进行建模和分析。 部分已有的研究工作是在假设卫星网络不存在处理延迟、传播延迟和丢包的情形下,采用SPN分析了卫星网络的性能。然而在实际的卫星网络中,丢包和延迟的影响不能被忽略。 为此,针对增加卫星网络建模和分析准确度的实际需求,本文主要考虑了卫星网络中消息传递的平均延时和通信过程中的丢包率,使用SPN进行建模分析,开展了如下独创性的工作: (1)证明了有关丢包率计算的定理; (2)考虑丢包因素影响,建立了LEO(低轨道)卫星网络在半双工通信模式和全双工通信模式下各自的GSPN(广义随机Petri网)模型,并利用SPNP(Stochastic Petri Net Package)6.0软件对这些模型的丢包率和平均延时加以性能分析; (3)建立了用于分析影响丢包率要素的LEO卫星网络的SRN(随机回报网)模型,并利用SPNP 6.0软件对丢包率的主要影响要素加以分析; (4)建立了LEO/MEO(中轨道)/GEO(同步地球轨道)三层卫星网络的GSPN模型,并使用SPNP 6.0软件进行了网络性能分析;通过OPNET(Optimum NetworkPerformance)仿真验证和分析,得出了诸如在高负载的情况下三层网络结构优于单层网络结构等一些新的结论。更多还原

【Abstract】 Based on the state-of-the-art of satellite network and its models, in this thesis we have analyzed the modeling methods for satellite network and the common methods for performance analysis of networks by comparison. The feasibility of physical experiment of satellite networks is slim because of the expensiveness of satellites and so on, making the combination of simulation with diverse mathematical models be the best choice for obtaining better performance. Modeling and simulation have been used t o s tudy t he p erformance o f s atellite n etworks, w hich o ffers u seful g uide a nd reference for the construction of satellite networks.At home and abroad, current tools for the performance evaluation of satellite networks are mainly based on queue theory and Markov process. Unfortunately queue model lacks the ability to model the relatively complicated structures in satellite networks such as parallelism, asynchronization and distribution of the message traffic in the satellite networks, as well as Markov process is hard to model the corresponding level of stochastic process. In contrast, SPN (Stochastic Petri Nets) is more powerful in the ability of description and also convenient to model the Markov process. Consequently, SPN is adopted to model and analyze satellite networks in this thesis.Some earlier investigations have been focused on a class of satellite networks usingSPN m odels b y assuming t hat p rocessing d elay, p ropagation d elay and packet 1 oss would be negligible. However, for actual satellite networks, the effect of packet loss and time delay can’t be neglected.The need for improving the quality of modeling and analyzing, we then improve the SPN modeling and analyzing of satellite networks by taking average time delay of message transmission and packet loss during communication into account.We have mainly engaged on following original works:(1)The relevant theory about the calculation of packet loss probability is proven.(2)Considering the effect of packet loss, GSPN (Generialized Stochastic Petri Nets) models have been constructed to analyze the average time delay and packet loss probability of LEO (low earth orbit) satellite networks under half-duplex mode and full-duplex mode with SPNP (Stochastic Petri Nets Package) 6.0, respectively.(3)Considering the effect of packet loss, a SRN (Stochastic Reward Net) model is constructed to analyze the main factors in packet loss probability of LEO (low earth orbit) satellite networks with SPNP 6.0.(4)The GSPN model of a triple-layered LEO/MEO (medium earth orbit)/GEO (geosychronous earth orbit) satellite network is constructed, and employed to evaluate the network performance with SPNP6.0 software package. The effectiveness of the results obtained by GSPN-based analysis is further verified in comparison with these by network simulation using OPNET (Optimum Network Performance), and some new results, such as the performance of the triple-layered network is better than that of the LEO one under high load, are derived thereby.更多还原