【作者】 王姝音；

【导师】 印桂生；

【作者基本信息】 哈尔滨工程大学 ， 计算机应用技术， 2013， 博士

【摘要】 随着Internet的计算环境的普及与不断发展，网构软件这一新的软件系统形态应运而生。区别于运行在静态、封闭、可控的环境之中的传统软件所具有的“有限自主性、固定封装性、交互单调性”等特征，网构软件具有“高度自治性、动态协同性、连续反应性、动态演化性和多目标性”。网构软件的出现为研究开放、动态环境下的软件理论与模型、软件技术与方法等提出了新的课题与挑战。从网构软件体系结构来看，网构软件由分布在网络环境下的自治、异构的软件实体和它们之间的协同方式所构成，软件实体通常由第三方所提供，其中封装了各类资源，它们独立自主的存在于网络的各个节点之上，为多用户提供服务，并且通过多种交互方式进行协同，网构软件不但可以满足用户不断变化的需求，还可以根据环境的变化进行动态的调整。海量、多样的软件实体和复杂、多变的协同方式使得开发可信的网构软件成为网构软件理论与技术方法研究中的重点与难点。在动态、难控的网络环境下实现智能可信的网构软件模型对网构软件的研究具有重要的价值和意义。因此，本文意在对网构软件的可信技术进行研究，在原有的网络安全技术的基础上增加信任行为可信的安全新方法，加强了对网络状态的动态处理，为实施自适应性、鲁棒的可信网构软件提供策略基础。本文的研究内容包括以下几个方面：(1)根据网构软件的基本特征与技术实现途径，提出一种支持Internet环境下软件服务实体实现在线协同演化的网构软件演化模型，对该模型的基本结构、核心机制及其具体实现过程给出详细的阐述和分析，并重点阐明该模型如何实现在线演化。该演化模型以运行时网构软件体系结构(RIA)为核心，RIA与运行系统之间存在一种因果关联关系，网构软件设计阶段的体系结构规约被具体化为内置于运行系统可编程的网构软件体系结构元模型(RIAMM)，通过一组自适应规则，可对RIAMM进行动态修改，进而深度控制运行系统实现动态演化；在演化模型基础上，对模型演化层中的信任机制进行研究，给出了选择软件实体加入系统的综合评判方法；最后对该模型服务层软件实体的技术实现进行了研究。为后续网构软件的可信模型研究提供了理论上的可行性和合理性保障。(2)以网构软件在线演化模型的整体架构为基础，研究网构软件系统中软件实体间的信任度量模型。针对现有信任度量模型粒度比较粗糙造成最终评估结果不够准确的问题，提出一种基于时间帧的多服务分级动态信任度量模型。模型中考虑到信任所具有的时衰特性，以时间帧作为动态信任模型的基本度量单位，针对不同服务所具有的权重不同，软件实体在交互协同过程中所基于的角色不同的问题，对信任度量值进行了细粒度化分析，同时考虑到个体经验差异对推荐可信度的影响，给出了基于经验因子的推荐可信度更新算法，弥补了个体由于自身经验不足造成的评估失准问题。除此之外，模型还给出了基于诱导有序加权算子的时间帧加权因子确立方法，使信任度量结果更加准确。利用该模型得出的信任度量结果将作为贝叶斯博弈模型的决策依据。(3)针对开放网络中网构软件实体的信任行为的复杂性，信任协作过程中体现出的类似人类社会的信任关系的特点，对实体节点的协作信任关系进行贝叶斯博弈建模。由于网构软件系统中实体节点在协作过程中掌握信任不完全，并不能确定与之协同的实体节点的类型，因此需要建立节点协作的贝叶斯博弈模型。模型中引入自然选择的概率，通过Harsanyi转换，将不完全信息博弈转换为完全但不完美信息博弈，并给出了信任协作贝叶斯博弈模型中的贝叶斯纳什均衡解，对博弈结果做出有效的分析与预测。在此基础上，给出了软件实体的贝叶斯信任决策过程，将信任度量模型得出的信任度量结果作为实体间信任协作的决策依据，指导软件实体进行策略选择，使得软件实体间的信任决策更加合理，以提高系统的整体效用值。(4)运用博弈论对网构软件系统可信模型中的激励机制进行研究。通过对贝叶斯博弈模型的分析发现，网络中的恶意软件实体在协作的过程中存在“搭便车”的行为，使得善意节点在与恶意节点的协作中陷入“囚徒困境”，为解决这一问题，本文建立软件实体信任协作的重复博弈模型，运用激励策略来激励软件实体节点自主的形成协作，并给出了不同激励策略成立的均衡边界条件。在此基础上，用演化博弈模型探讨软件实体节点的种群演化趋势，通过建立复制者动态方程，找到符合网络进化的唯一演化稳定策略，并根据此探讨促使网构软件系统向演化稳定策略收敛的激励惩罚机制。更多还原

【Abstract】 With the increasing popularity and development of the computing environment of theInternet, the Internetware, a new form of the software system, comes into being. Unlike thefeatures (such as the limited autonomy, fixed encapsulation and interactive monotonicity) ofthe traditional software which runs in a static, closed and controlled environment, theInternetware is of high degree of autonomy, dynamic synergy, continuous reactivity, dynamicevolution and multi-objective. The presence of the Internetware has brought new issues andchallenges to the research on, for example, the software theory and model under the open anddynamic environment, the software technology and the method. Judging from the systemarchitecture of the Internetware, the Internetware is comprised of software entities and theircollaborative method. These software entities are autonomous, heterogeneous and distributedin the network environment, and are usually provided by the third party. They encapsulatevarious resources and reside independently on individual nodes, offering services to multipleclients and collaborating in different interactive ways. The Internetware can not only meet theever-changing client requirements, but also make dynamic adjustment in accordance with thechanges to the environment. The massive and diverse software entities and the complex andever-changing collaboration manners have made the development of the creditableInternetware become the important and difficult aspect of the study the Internet theory andtechnology. It has important values and significance to study the realisation of the intelligentand creditable Internetware model in the dynamic and difficult-to-control environment. Assuch, this paper intends to study the Internetware creditability techniques. It proposes a newsecurity technique that adds the trust behaviour to the existing network security techniques,which has enhanced the dynamic processing of the network status and provided strategicfoundations to implement self-adaptive, robust and creditable Internetware. The researchfocus of this paper includes the following:(1) Based on the fundamental features the technique implantation methods, to propose anInternetware creditable evolution model, within which the software service entities achieveonline collaboration evolution in the Internet environment. To detail and analyse and the basicstructure of the model, the core mechanism and its detailed implementation process, and in particular, to detail how the model achieves the online evolution. The evolution model isbased on the runtime Internetware architecture (RIA). The RIA and the running systems havethe casual connections. The rules on the Internetware architecture during the design phase hasbeen detailed as the runtime Internetware architecture meta model (RIAMM), which isembedded in the running system and programmable. RIAMM can be modified dynamicallyvia a set of self-adaptive rules in order to achieve the dynamic evolution by controlling therunning systems at a deeper level. Based on the evolution model, to study the trust mechanismin the evolution layers and come up with a comprehensive evaluation algorithm to selectsoftware entities to join the system. In the end, to study the implementation of the softwareentities techniques in order to support the study the feasibility and the rationale of thecreditability model of subsequent Internetware at the theoretical level.(2) Based on the over architecture of the Internetware online evolution model, to studythe creditability measurement model of the Internetware. With regards to the problem that thefinal evaluation result is inaccurate due to the credibility measurement model granularitybeing rough, to propose a dynamic creditability measurement model based on multiple taskclassification. Based on the timeframe based dynamic creditability model, the model allocatesdifferent weights further to the classification different tasks and calculates the creditabilityvalues of the corresponding node, which has contributed to the granularity of the evaluationresult. In the meanwhile, considering the impacts on the recommended creditability and thefeedback credibility due to the difference of the individual experience, to introduce experiencefactors into the calculation of the feedback creditability, which has make up the evaluationinaccuracy problem caused by the insufficient experience of individuals. In this model, anentity node not only needs to evaluate the creditability of the nodes that it collaborates, butalso needs to evaluate itself in order to obtain the creditability information that other entitynodes calculate against itself. The purpose is to use the both evaluation results as the decisioncriteria for Bayesian game modeling.(3) With regards to the complicity of the creditability of the software entities within theopen Internetware and the feature of the trusting relationship through the trustworthycollaboration process which is similar to the one in the social society, to Bayesian gamemodeling against the trusting relationship between entity nodes. Since the Internetware entitynodes obtain incomplete information during the collaboration process and are unable to confirm the type of the entity nodes that they collaborate with, it is necessary to establish theBayesian game model of the node collaboration. This introduces the probability of natureselection and converts the game of incomplete information into the game of complete butimperfect information via the Harsanyi conversion. The focus is to use the creditabilitymeasurement information from the second part as the base for decision-making, guide thenode during the game, analyse and predict the game result effectively.(4) To study the incentive mechanism of the Internetware credibility model, using thegame theory. The analysis of the Bayesian game model shows that the selfish software entitieswithin the network have the“free riding” behaviour during the process collaboration witheach other, which traps the collaboration between selfish nodes to "the prisoner dilemma". Tosolve this problem, this paper establishes the repetitive game model between the selfishsoftware entities and looks for the proper strategy to encourage the software entitiescollaborate autonomously. Based on this, to discuss the evolution trend of the software entitypopulation using the evolution game model, to find the only stable evolution strategy inaccordance with the network evolution through the establishment of “Replicator dynamicequation”, and to discuss the incentive punishment mechanism in order to assist theInternetware in moving toward the stable evolution strategy.更多还原