【作者】 罗贞；

【导师】 李胜利；

【作者基本信息】 华中科技大学 ， 计算机软件与理论， 2004， 硕士

【摘要】 分布式系统的可靠性不仅取决于它的资源(如程序和数据文件)的分布,而且取决于它的通讯链路和节点的可靠性。分布式实时软件还应考虑时间约束因子。在对传统的分布式软件可靠性评估理论的分析和研究的基础上,提出了一种增加时间约束因子的实时分布式软件可靠性评估模型。它对文件生成树中的数据文件传输时间进行评测,确定文件生成树的状态,建立马尔可夫模型,利用因式分解理论计算出分布式程序的可靠性,从而得出整个实时分布式软件的可靠性。为了加快评估过程,使用了几种常用的可靠性保留的精简算法来简化分布式系统,减小马尔可夫状态空间。设计和实现了一种得到分布式程序的所有文件生成树的递归算法。文件传输路径生成算法借用了传统的最小文件生成树生成算法中节点扩展的思想,并且判断每一个新生成的文件传输路径是否已在文件传输路径集合中,如果是则丢弃,这样就避免了生成相同的文件传输路径,同时确定了数据文件的唯一分布。评估文件传输路径的响应时间时,在直线计算方法的基础上,添加了分支路径计算方法。建立马尔可夫模型并由马尔可夫状态转移概率矩阵计算出程序的可靠度。实时分布式软件可靠性评估系统包括节点机模拟器、分布式软件模拟器和可靠性评估器。节点机模拟器和分布式软件模拟器共同模拟被测的分布式软件,辅助测试过程,实现程序的随机调度和文件的随机分配。可靠性评估器采用从配置文件读入分布式系统的拓扑结构的方法,可测试具有任何拓扑结构(如直线、星形、环形、网状)的分布式系统的可靠性。通过改变时间约束因子和链路的通讯能力等参数,对系统进行了大量的测试,发现不同的分布式系统配置和参数对分布式软件可靠性的影响特征。更多还原

【Abstract】 The reliability of distributed computing system depends not only on the reliability of communication links and nodes, but also on the resources distribution, including programs and data files.Reliability models so far proposed for distributed software reliability evaluation can’t capture the effect of real-time constraints. This proposal presents an approach for the reliability performance analysis of distributed software addressed on the real-time constraint issue. With time-constraint information and the execution time of each File Spanning Tree (FST), the corresponding Markov state space can be defined for reliability computation. To speed up the evaluation process and reduce the size of the Markov state space, several dynamic reliability-preserving reduction techniques are developed. A realizable recursive algorism is proposed to search all the File Spanning Trees for a given program in the distributed system. To search all the File Transmission Paths (FTP) for each FST, we make use of the node-add method used in Minimum File Spanning Tree (MFST) and check if the FTP set includes the same tree before an FTP added. In the evaluation of file transmission time, based on the algorism for linear transmission path, we take the branched path into account to make the algorism perfect. We also present how to compute the reliability of a program based on factoring method and Markov model.The Time-Constrained Distributed Software Reliability (TCDSR) evaluation system is composed of Node Emulator (NE), Distributed System Emulator (DSE) and Reliability Evaluator (RE). The NE and DSE, which dispatch the programs and files randomly, together emulate the distributed software. The RE gets the information of all the nodes to work out the metrics. The test result illuminates that the TCDSR system can be applied to distributed system with any topology, such as line, star, ring and net topology. By changing the parameters, such as time-constrainat and the reliability of the links, the test result can help the distributed system engineer to optimize the performance of the system and software.更多还原