【作者】 王海鹏；

【导师】 周兴社；

【作者基本信息】 西北工业大学 ， 计算机应用技术， 2006， 博士

【摘要】 普适计算是下一代计算模式发展的主要方向之一，普适资源管理技术是普适计算研究的重要课题。2003年IEEE Pervasive Computing期刊主编提出了“建立随机意义下普适资源管理的概念”。本文研究的是随机意义下、面向用户的适应性普适资源管理技术，目标是提供随时随地的资源可用性。本文结合国家自然科学基金课题“下一代计算机体系结构和系统软件”和国家自然科学基金课题“面向复杂分布实时应用的自适应资源管理模型与机制研究”，从理论和实践两个方面探讨了随机意义下的、面向用户的、以可用性为目标的适应性普适资源管理问题。本文完成的主要工作成果包括以下几个方面。●针对普适资源强异构、高度动态和随时移动的特点，本文建立了按设备、任务和用户，逐层管理的三层适应性活动资源空间模型ARS，实现了可平滑过渡的层次化普适资源自适应管理机制。●针对现有资源描述方法缺乏对资源随机信息表示的支持，应用概率模型的思想，提出了一个基于OWL-S扩展的普适资源描述方法OWL-SP，增加了对随机信息部分的表示，扩展部分包括相应的顶层本体、属性和约束。OWL-SP较为全面地支持了普适资源描述的服务化、语义化、用户化和随机化需求，形成了一个较为完整的普适资源描述。●资源分配是资源管理系统的重要组成部分，现有研究大多考虑确定性情况下的动态资源分配问题。本文针对普适环境下用户任务随机演化的特点，提出了一个两阶段的普适资源随机预分配算法，以解决随机任务意义下的普适资源优化分配问题。建立了一种基于随机规划的普适资源随机分配模型，以支持随机意义下的资源优化分配。实验分析表明，该方案能够保障概率条件下的用户任务执行，并且能够有效降低用户任务的资源等待延迟，提高用户服务质量。●针对资源可用性随机变化情况下，需要执行资源适应性调整的问题，本文建立了一个基于控制马尔可夫链的资源级适应性调整模型，模型能够同时考虑资源管理行为和资源状态两者对资源可用性的影响，最终提供一个优化的资源适应性调整策略。●针对现有研究大多从定性角度分析普适资源的可用性，本文建立了基于马尔可夫重生过程的普适资源可用性定量分析模型，包括用户会话可用性分析模型和用户请求可用性分析模型。前者针对用户会话特点，放松了用户状态驻留时间须为指数分布的约束，扩展到任意分布情况，因此包含了离散和连续马氏链的特例情况，是一类更为泛化的可用性分析模型；后者针对用户请求特点，放松了在用户行为重生点之间资源状态不变的约束，并且支持任意分布，因此能够更加准确和细化地描述用户请求级上的可用性变化。●为了使定量计算更有针对性，本文提出了三类普适资源可用性度量指标，包括单步用户资源同步预留可用性度量TURS+、用户会话可用性度量和用户请求可用性度量。其中，TURS+能够刻画普适资源可用性中用户与资源的同步关系和预留特征：用户会话和用户请求可用性度量能够分别从全局会话和局部请求层次描述资源的可用性变化。实验分析表明，TURS+能够更为严格地保障用户会话和请求的执行，并且其预留性质可以为普适资源预管理活动提供参考；用户会话和用户请求可用性度量能够为系统的设计和选择提供指导。●在以上研究成果的基础上，本文以军事应用为例，设计并实现了普适计算环境原型系统-UbiPresn，探讨了普适资源管理技术的应用问题，验证了相关技术的有效性。更多还原

【Abstract】 As a main computing paradigm in the future, the pervasive computing presents dis-tinct challenges on the pervasive resource management. In 2003, the Editor in Chief of IEEE Pervasive Computing proposed the new concept of stochastic pervasive resource management.Sponsored by the NSFC project of "Future Computer Architecture and System Soft-ware" and the NSFC project of "Adaptive Resource Management Model and Mechanism for Complex Distributed Real-Time Applications", this dissertation discusses the stochas-tic, user-oriented, and adaptive resource management with the aim of providing resource availability anytime and anywhere, from both points of theoretical and practical views. The main points of this dissertation are described as follows:●Based on the distinct characteristics of pervasive resource, strong heterogeneity, high dynamism, and mobility in anytime, this dissertation develops a model of ac-tive resource space, which is capable of providing three layers of adaptation in device, task, and user. Based on this model, it is able to provide seamless adaptive pervasive resource management across different layers.●A new approach, called OWL-SR is proposed to describe pervasive resource with uncertainty. It is achieved by extending OWL-S with the expression power of stochastic information of resource. The extension includes the related upper on-tology, attribute, and restriction. The new approach makes the pervasive resource description service-encapsulated, semantic-based, user-oriented, and stochastic-enabled.●Resource allocation is an important issue of resource management. Currently most researches focus on the dynamic resource allocation under deterministic conditions. To address the challenges of resource allocation under stochastic tasks evolution, an algorithm of stochastic resource pre-allocation with two stages is proposed. It includes a core model of stochastic allocation of resource based on stochastic pro- gramming. The experiments show that this approach supports the execution of tasks in the probabilistic constrictions, which can decrease the delay occurred dur- ing the resource allocation effectively, and improve the quality of service perceived by users finally.●Based on the fact of stochastic change of resource availability, it is necessary to carry out adaptive adjustment for related resource. A controlled markov chains-based model is developed to achieve such adaptive adjustment, which considers the effects on resource availability by two factors, the action of resource management and the resource states. Finally an optimal policy for adaptive adjustment can be produced.●Two MRGP-based quantitative analysis models for the pervasive resource avail-ability are developed to concern the user session and the user request respectively. The former takes the arbitrary distribution, describing the sojourn time of each user state, into consideration, so it is a more general model, covering the particular cases of DTMC and CTMC respectively. The latter takes the change of resource states, between two regeneration epochs of user behaviors, into consideration, and also supports the arbitrary distribution. As a result, it provides more accurate and de-tailed description of resource availability for user request.●To support the measure of resource availability in special pervasive computing ap-plications, three measures, TURS＋, user session, and user request measures, are defined. The TURS＋is capable of describing the characteristic of synchronized association between user and resource with the reservation of availability. The user session and user request measures are able to describe the availability from the global session and local request respectively. Experiments show that the TURS＋supports the execution of user session and user request more strictly, and its reser- vation feature can provide reference for the pervasive resource management. The user session and user request measures are able to provide certain guide for system design and selection.●A prototype for ubiquitous computing environment, called UbiPresn, is designed and implemented, which verified the effectiveness of the above proposed pervasive resource management techniques.更多还原