【作者】 冯光升；

【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， 2009， 博士

【摘要】 随着网络技术的不断发展和应用范围的不断扩大,网络系统越来越复杂,一些新兴通信应用不断涌现,致使网络性能及端到端系统性能得不到保障。受限于传统网络层次化结构的限制,当前网络元素不能感知其它网络元素的各种行为及环境状态,更多地依靠人工干预,对复杂环境的适应能力明显不足,难以满足人们所希望得到的网络服务质量(Quality of Service,QoS)。认知网络(Cognitive Networks,CN)正是在此需求背景下而产生,并成为网络工作者面临的一个重要课题。所谓认知网络就是网络系统具有感知环境变化和网络状态的能力,根据整体目标及端到端目标,通过适当的学习机制,利用感知的环境信息和网络状态信息,实时动态地调整网络配置,智能地适应环境变化并能指导未来的自主决策。目前,认知网络作为下一代网络(Next Generation Network,NGN)的核心内容和发展趋势,其研究正处于起步阶段,而认知网络应用环境的复杂程度远超过开发者在系统设计初期所预想的程度,几乎所有国家在部署下一代网络实施要求时,都提出了要解决认知网络端到端QoS问题。如何建立保障通信的框架结构并提供高效的网络QoS以保证认知网络端到端性能成为认知网络的核心内容和研究热点。本文首先从系统框架这一宏观角度出发,结合认知理念,提出面向认知网络的自适应QoS框架,为后续工作的开展打下基础;然后从边界网络和主干网络两个角度分别对认知网络QoS数据的收集和分析进行研究,为下一步的QoS动态自配置提供数据支持;最后使用效用函数对QoS进行表示和区分,对QoS动态自配置进行研究。本文从体系架构到数据处理分析再到系统配置,构建一条保障端到端QoS的途径,以满足用户对认知网络QoS的需求。主要研究内容组织如下:首先,本文提出了具有跨层感知能力的认知网络自适应QoS框架。考虑到下一代网络的业务需求,结合认知网络QoS设计理念,突破传统网络层次结构,提出一种自适应的QoS框架结构,该结构自下而上分为“系统层—服务层—用户层”三个逻辑层次。通过在每个层次上部署QoS管理单元来实现网络元素的互操作,建立层间QoS管理单元独立通道来实现系统的跨层感知,以实现一体化的QoS控制。形式化分析结果表明,本文设计的认知网络QoS框架能够体现认知网络跨层感知、自我管理和动态感知等方面设计需求。其次,针对端到端的QoS保障要求,本文从边界网络和主干网络两个方面对网络QoS数据进行采集和分析。针对边界网络的QoS感知问题,提出了“两阶段”的数据处理方法即基于DS证据理论的QoS数据处理(Data Preprocess,DPP)和动态权重提升(Dynamic AdaBoost,D-AdaBoost)相结合的QoS数据预处理方法,将感知到的边界网络数据进行融合预处理,DPP无法处理的数据将由D-AdaBoost提升方法对“难分”数据进行二次分类,达到根据多层数据实时感知网络QoS状态的目的。实验结果表明,“两阶段”的数据处理方法不仅能保证分类的精确度,去除不确定性噪声数据带来的不利影响,有效的避免DPP中的证据冲突,而且能提高QoS数据分类能力。针对主干网络的QoS感知问题,提出了基于中值树的IP流聚合方法。通过对主干网络上的IP数据流特点进行分析,从“聚”和“合”两个角度对主干网络IP流进行分析,并综合考虑主干网络IP流的统计特征,给出了IP流特征动态合并的方法。仿真实验结果表明,此方法不仅能在离线情况下高时效地处理大规模样本,而且在在线时也能通过统计特征的动态合并准确实时地反映主干网络IP流的特点,为主干网络实时QoS保证提供参考。最后,本文提出了基于效用函数的QoS动态自配置方法。充分考虑了下一代网络的认知特性和认知网络的自我管理能力,提出一种面向认知网络的用户QoS动态自配置方法,使用效用函数对用户QoS优先级进行表示和区分,并采用中断策略对用户QoS优先级进行动态修正,解决网络阻塞状态下对用户QoS支持不足的问题,使网络用户群体在当前的网络条件下获得最优的QoS。仿真实验结果表明,该方法可以在一定程度上减轻瓶颈链路的拥塞状态,降低数据包的丢失率,在发生链路堵塞的情况下能有效地保持用户满意度。更多还原

【Abstract】 With the rapid development of networking technologies, its application scope has been extended every corner of the world. Specifically, network systems have become extremely complicated with the appearance of new communication technologies, which has resulted in the unreliability of network and end-to-end performances. Those current network elements could not be awareness of the action of others and the environment state due to the constraint of traditional network architecture. As a result, the adaptaion ability of network elements for complex environment is inadequate and the quality of service (QoS) can not be guaranteed. Therefore, the concept of cognitive networks (CN) is generated under this background and become an important topic for the researchers in network fields.The so called cognitive networks are the specific network systems that have ability of environment-awareness and network state-awareness. Based on awared information, the whole system targets and end-to-end targets, cognitive networks can utilize proper learning mechanism to adjust network configuration, adapt to environmental change intelligently and guide future autonomous decision-making.Cognitive networks are considered as the core and development trend of the next generation network (NGN) in commom, which now are on its infant stage, and the application complexity of cognitive network is beyond the original understanding of system designers. The end-to-end QoS problem in cognitive networks is proposed when NGN is deploied in most of the counties. How to construct the architecture and provide the effective network QoS have become the key topics in the field of cognitive networks.In this paper, we first propose a self-adaptive QoS framework of cognitve network combined with the cognitve philosophy. As a data support for next dynamic self-configuration, we then investigate the bordernet and backbone QoS data collection and analysis. Finally, QoS metrics are distinguished by utility function and the dynamic self-configuration mechanism is proposed. In this paper, we propose an approach to guarantee the end-to-end QoS in order to meet the users’need. The main contents are organized as follows.Firstly, a cognitive network QoS framework is proposed, which is ability of cross-layer awareness. Considering the demand of NGN as well as design philosophy of cognitive network QoS, a self-adaptive QoS framework is proposed which is devided into three logic layer called as“system layer-service layer-user layer”from bottom to top. QoS management elements on every layer are deployed to achive the mutual operation, and the channels are constructed to obtain the cross layer awareness and integrative QoS control. Formal analysis result shows that this proposed QoS framework meets the need of cognitive networks demand which can demonstrate philosophy of cross layer design, self-management and dynamic awareness.Secondly, according to the requirements of end-to-end QoS, a data preprocessing approach is proposed in the views of bordernet and backbone.In order to solve the bordernet QoS-awareness, a“two step”method are proposed. In the first step, DPP (Data Preprocessing) based on DS evidence theory is utilized to fusion network QoS data and classify them in the meanwhile. In the second step, the D-AdaBoost (Dynamic AdaBoost) is utilized to classify the remained QoS data by the first step.The experimental results show that our method can assure the precision of classification, eliminate the influence of uncertain noisy data and avoid the evidence conflict.In order to solve the backbone QoS-awareness, IP flow aggregation method based on median tree is proposed in this paper. By analyzing the IP flow statistical characteristics of backbone networks, the propoed method has a capability of analyzing backbone IP flows in the views of“cluster”and“combination”. The simulation experiments show that IP flow aggregation method can handle the large scale data effectively offline. Moreover, our method can demonstrate IP flow characteristics of backbone networks accurately by statistics characteristics combination.Lastly, we propose a QoS dynamic self-configuration method based on utility function. Taking cognitive characteristics and self-management ability into consideration, the utility function is used to distinguish QoS priority and adopt interruption policy to dynamically modify the QoS priority which can reinforce the users’QoS support when the network is congested. The experiment results show that our method can alleviate the congestiong on the bottleneck links to some extent, reduce the data packets lost and maintain the user satisfaction when bottleneck links are congested.更多还原