【作者】 张海涛；

【导师】 王慧强；

【作者基本信息】 哈尔滨工程大学 ， 计算机应用技术， 2010， 博士

【摘要】 随着计算机技术的快速发展和广泛应用,计算机软件和硬件设备高度集成,传统技术无法满足人们对系统可靠性、安全性等需求,迫切需要新的理论和方法去解决软件管理和安全危机,因此自律计算的概念应运而生。自律计算能够降低人工干预的频率和复杂度,增强系统的可靠性和安全性。但目前自律计算研究还处于初级阶段,许多关键性问题还没有解决,尤其是缺乏自律计算系统的自律成熟度评估研究,制约着自律计算的进化发展。本课题针对自律计算缺乏评估指标和度量方法,不能进行系统自律性评估和量化分析等问题,将可信性与自律性相结合,以面向服务的视角对系统服务的自律可信性进行评估指标和建模方法研究,实现系统服务自律可信性的定量描述和分析问题,本文的主要研究内容如下:首先从面向服务的角度,建立基于多Agent的自律计算系统评估分析模型,并应用Web服务和XML对自律计算系统的状态和行为进行建模。面向服务的思想只关注自律计算系统的服务特性,不关注系统具体实现的技术,为自律计算系统评估建立统一的评估标准。实验结果显示,该分析模型能够准确地反映自律计算系统服务的关键属性和自律特征,本章为自律可信性的建模及量化分析提供指导作用。其次,依据自律计算系统评估模型中自律单元与服务之间的关系,从服务请求、服务过程、服务响应三个方面对自律可信性进行量化指标分析,提取影响自律服务可信性的综合指标,建立分层的自律可信性评估体系。经过实例验证表明,该评估体系对系统服务进行分析,可应用于不同自律计算系统的自律可信性评估,本章为进行系统级自律可信性分析提供了合理依据。再次,提出基于聚类分析的HMM自律可信性建模方法,将系统的自律行为和服务状态映射为隐马尔科夫的双重随机过程,对系统状态进行聚类分析,提取有效的自律状态进行建模。提高模型对目标系统的识别能力,并采用信息熵对模型进行参数优化,降低随机参数的选取,优化模型精度。实验结果显示,该方法可以准确地反映自律可信性各要素之间的逻辑关系和变化,提高了评估分析的准确性,是进行量化分析的重要基础。最后,研究基于支持向量机(SVM)的系统自律可信性预测方法,将系统服务运行数据作为SVM的输入样本进行分类,输出不同服务等级的概率,实现了服务自律可信性的量化分析预测。该方法在小样本信息量较少,数据模糊的条件下,克服了原始评估数据非线性和离散性等缺陷,实现了自律可信性的量化分析,仿真实验结果显示预测的精度较高,为优化和指导自律计算系统的设计提供帮助。更多还原

【Abstract】 With the rapid development and wide applications of computer technology, and highly integration of software and hardware devices, traditional technologies cannot meet the need of the reliability and security any longer. New theories and methods are expected , which Autonomic Computing(AC)theory is proposed to resolve the issues of computer management and safety crisis,which the frequencies and complexities of manually intervention are reduced. Then the reliability and security of system is improved. However, AC research is still in infancy, many crucial technologies have not been solved, especially in autonomic maturity research of AC system evaluation. Evolution process of AC is limited.At present, there are few criterion and measurable methods to analyze quantitatively autonomicity of AC system.In this dissertation, combining dependability with autonomicity, the work is mainly focused on the autonomic-dependability indexes and modeling of system service based on service-oriented. The research aims to solve issues of qualitative description and analysis of AC service evaluation, and the main research contents are organized as follows:First, a multi-agent analysis model of AC evaluation based on service-oriented is established. States and behaviors of system are described by Web service and XML technology to modeling.Service-oriented idea focuses only on the service characteristic of AC system instead of special technologies. A general standard of AC system evaluation is set up. Experiment results indicate that it can accurately reflect service key properties and autonomic characters of AC system. The multi-agent model provides the guiding in model and quantitative analysis of AC system.Second, according to the relationship between autonomic unit and service of AC system evaluation model, autonomic-dependability is quantitative analyzed from three respects of service requests, service processes and service responses, comprehensive indexes which has an important impact to autonomic services are extracted. Then a hierarchical evaluation system for autonomic-dependability is established. Application in specific case indicates that the research can be used to analysis and evaluation in different AC systems. The method provides a reasonable basis for autonomic-dependability analysis.Third, HMM autonomic-dependability evaluation method based clustering analysis is proposed. Autonmoic behaviors and service states are extracted into HMM. Effective autonoimc states are extracted from system states by clustering algorithm, in order to improve model ability to the recognition of target system.Then, information entropy is applied to optimize the model parameter selection which improves model accuracy. Experiment results show that the model can accurately reflect the logic relation and transformation for autonomic-dependability elements. It is an important basis for quantitative analysis.Finally, a forecasting method for system autonomic-dependability based on Support Vector Machine (SVM) is proposed. The autonomic-dependability is forecasted by regarding the system service data as SVM input and the service level probability as output. The method overcomes drawbacks of the origin evaluation data such as nonlinearity, fuzzy, uncertainty and discreteness. Quantitative analysis for autonomic-dependability is realized. Experiment results show that has a higher forecasting precision. The research provides to optimize the AC system design.更多还原