智慧协同网络下基于需求感知的资源适配方法研究

发布时间：2018-01-17 14:28

本文关键词：智慧协同网络下基于需求感知的资源适配方法研究　出处：《北京交通大学》2017年博士论文　论文类型：学位论文

【摘要】：随着用户规模的扩大和应用种类的增加,互联网已经成为一个承载数据、多媒体等种类繁多、特征各异、海量应用的综合业务网络。然而,传统TCP/IP网络体系仅提供尽力而为且粗粒度的服务,而不承载用户具体需求和业务特征,使得网络难以根据用户具体需求和业务特征对网络资源进行智能分配,导致了用户体验不佳、资源利用低、服务质量差等弊端。为此,本文以国家973项目"智慧协同网络理论基础研究"为背景,从网络如何感知用户和服务需求、如何利用用户和服务需求等问题出发,研究基于需求感知的资源适配方法。论文的主要研究工作与创新点如下:(1)针对网络如何感知用户和服务需求的问题,提出了一种需求感知通用基础框架,并在原型系统中验证了其可行性。首先,分析了在网络体系架构中提供用户和服务需求感知能力的必要性,并讨论了架构设计时面临的主要挑战。在此基础上,提出利用XML语言对用户和服务需求进行描述,在报头中携带XML语言描述的需求信息,并利用控制器读取用户和服务需求。再次,在智慧协同网络原型系统中对提出的需求感知通用基础框架进行了实现,并以感知服务内容大小这一服务需求为例,结合域内路由适配机制,从可行性和大流探测两方面进行了测试评估。实验结果表明,该框架能够实现对需求信息的合理感知,且具有良好的架构兼容性,显著减轻了数据流探测和网络资源优化等方面的负担开销。(2)针对如何利用用户需求提升用户体验的问题,提出了一种感知用户需求的个性化服务选择模型和自适应分级传输机制,可实现面向用户的差异化服务适配并提升链路带宽利用率。针对服务获取的解析阶段,建立了基于用户偏好需求的服务选择统计概率模型,并运用蒙特卡洛仿真对服务选择过程进行随机抽样。从理论和仿真两个角度分析不同场景下用户偏好对服务选择的影响,得到了服从负指数分布的服务选择变化规律。针对服务获取的传输阶段,提出了基于用户级别感知的自适应分级传输机制,通过引入用户级别需求与用户级别管理器模块,设计了一种基于用户级别的链路带宽适配方案。仿真结果表明,该方案实现面向用户的差异化分级传输质量,并提升链路带宽利用率25%以上。(3)针对网络在已知服务需求时的资源动态适配问题,提出了一种基于期限需求的低开销数据流传输控制机制(DSLO),可降低数据中心网络无效期限流所引发的网络开销。DSLO提出了最大化期限满足率和最小化无效流开销的总体设计思路,以及接受、预测和中止三种资源适配控制功能。在此基础上,通过发送端驱动的平均传输速率监测,实现了一种基于发送速率的无效流预测方法,并对该方案进行鲁棒性分析。与此同时,在交换机端建立了与降速倍数相关的无效期限流概率预测模型,并提出了一种基于预测概率的自适应预测方法。最后,在典型数据中心网络拓扑下,针对包括DSLO在内的不同期限流传输机制,从网络开销、期限满足率、参数选择合理性及机制鲁棒性四个方面进行仿真测试与评估。仿真结果表明,DSLO机制能够实现对无效期限数据流的合理预判,最优情况将网络开销从10%削减至1%以内并维持90%以上的期限满足率。(4)针对服务需求感知如何有效提升服务质量的问题,提出了一种基于期限需求的数据流调度机制(FBDS),实现了基于期限约束的集中式停等调度决策,可有效提升数据中心网络高并发流情况下的期限满足率。FBDS机制的核心思想是采用集中式的控制方案,推迟一些数据流至期限到达时刻完成,通过预留传输空间的方式,缓解高并发流情况下的期限流失效问题。FBDS机制通过引入传输调度块和停等操作,设计了期限流的停等调度算法,并从调度块独立性和停等机制合理性两个方面对停等调度进行理论推导验证。最后从数据流平均完成时间、期限满足率、停等程度三个方面对FBDS机制进行仿真评估测试。仿真结果表明,FBDS机制在高并发流情况下,可以将期限满足率从最差的45%左右提升至85%以上。
[Abstract]:With the increasing expansion of the scale of users and applications, the Internet has become a bearer of multimedia data, variety, different characteristics, massive application of integrated services network. However, the traditional TCP/IP network system only provides best effort and coarse-grained services, without bearing the specific needs of users and business characteristics, making the network according to the cyber source intelligent distribution of the specific needs of users and business characteristics, leads to a poor user experience, low resource utilization, disadvantages of poor quality of service. Therefore, based on the background of national 973 project "intelligent cooperative network theory", from the network to the user perception and service needs, of the problem of how to use the user and service needs the research, demand aware resource adaptation based approach. The main research work and innovations are as follows: (1) according to the network to users and Service Perception The problem of demand, proposes a general demand perception framework, and the feasibility is verified in the prototype system. Firstly, analyzes the necessity of providing users and service demand perception ability in the network architecture, and discusses the main challenges in architecture design. Based on this, proposed carries on the description to the user demand for services and the use of XML language, with XML language to describe the demand information in the header, and read the user and the service demand by the controller. Again, in the wisdom of the collaborative network prototype system on the basis of general demand perception framework is presented, and the perceived service content size of this service needs for example, binding domain in the routing adaptation mechanism, tested and evaluated from two aspects and the feasibility of flow detection. The experimental results show that the framework can achieve a reasonable perception of demand information, and has a good Architecture compatibility, significantly reduces the data flow detection and cyber source optimization etc. the burden of expenses. (2) on how to use the user needs to enhance the user experience of the problem, put forward the personalized service of a perception of the choice of user demand model and adaptive hierarchical transmission mechanism, can realize the differential service oriented user adaptation and enhance the utilization of the rate of link bandwidth. According to the analytical phase of service access, a user preference demand service selection model based on probability statistics and Monte Carlo simulation, random sampling of service selection process. Analysis of the impact of user preference under different scenarios of service selection from two aspects of theory and simulation, the exponential distribution of service selection changes in the law. For the transmission service acquisition stage, proposed an adaptive hierarchical transmission mechanism based on the perception of the user level, through the introduction of The user level requirements and user level management module, design a bandwidth adaptation scheme based on user level. The simulation results show that this scheme is different user oriented classification transmission quality, and enhance the link bandwidth utilization rate of more than 25%. (3) according to the adaptation problem of dynamic network resources in known service requirements that presents a low overhead data period based on the demand of the stream transmission control mechanism (DSLO), can reduce the network overhead of.DSLO data center network flow caused by the invalid period puts forward the overall design idea, the maximum period of satisfaction rate and minimize the invalid stream overhead and acceptance, forecast and suspension of three resource adaptation control function. On this basis, the average transmission rate monitoring driven by the transmitter, realizes an invalid flow forecasting method based on transmission rate, and the scheme for robustness analysis. At the same time, in the end established a valid switch associated with multiple flow deceleration probability prediction model, and proposes an adaptive prediction method based on probability prediction. Finally, in a typical data center network, according to the different period including DSLO transmission mechanism, from the network overhead, meet the deadline rate parameter selection in four aspects of rationality and mechanism of robustness test and evaluate the simulation. The simulation results show that the DSLO mechanism can the reasonable pre data stream is valid to achieve optimal conditions will be sentenced, network cost reduced from 10% to 1% and to maintain more than 90% of the time to meet rate. (4) for the service perception of how effective demand to improve the quality of service, this paper proposes a scheduling mechanism based on the data demand period (FBDS), the deadline constraint of centralized stop scheduling decision based on data center network can effectively improve Complex high concurrent flow conditions of time limit and meet the core idea of.FBDS mechanism is rate control scheme using centralized, delayed data stream to the period of the arrival time of completion, by reserving transmission space, alleviate the high concurrent flow under flow period.FBDS failure mechanism by introducing the transmission scheduling block and stop operation, design the term flow stop scheduling algorithm, and the scheduling block independence and stop mechanism rationality of two aspects of the stop scheduling theory derivation and verification. Finally, the stream of data from the average completion time, deadline satisfaction, stop of three aspects of simulation and evaluation test on the FBDS mechanism. The simulation results show that the FBDS mechanism in high concurrent flow, can meet the worst period rate from about 45% to more than 85%.

【学位授予单位】：北京交通大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP393.0

【相似文献】