命名数据网络中基于内容和端口状态的转发策略研究

发布时间：2018-08-10 18:11

【摘要】：随着互联网中新型应用的不断出现和人们对网络服务需求的不断增加,基于TCP/IP传统互联网架构在移动性、扩展性等方面表现出各种各样的不适应性,尤其是在当前持续增加的以数字媒体等应用为代表的新兴分发服务模式中。为了解决目前互联网架构与新型应用之间的不适应性,国内外研究学者提出了重新设计未来互联网架构的研究思路。作为未来互联网架构体系中的重要成员之一,命名数据网络(Named Data Networking,NDN)因其具有完整的路由协议框架和层次化的命名规则而备受关注。在NDN中,网络节点不单单提供转发功能,同时还提供存储能力。路由节点中缓存的添加不仅可以避免网络中数据的重复传输,还为数据的传输添加固有的多路径特性。转发策略的研究对于充分利用固有的多路径特性来实现数据的高效转发具有重要的研究意义。本文的主要工作如下:1、针对当前NDN转发策略中包的乱序问题,本文首先提出了一种基于MDP(Markov Decision Processes)的转发策略。在该策略中,单个节点中数据转发决策过程Markov决策过程模型被建立,然后通过对MDP模型的求解,最优的转发策略被获取。在MDP建模过程中,各个端口不同类别流的个数作为系统状态的设置实现网络内容分布和端口状态的联合感知。2、针对当前NDN转发策略中网络端口和内容分布共同感知的不足,本文又提出了一种基于决策树的转发策略。该策略将整个转发流程划分为离线学习和在线转发两部分进行,在线转发过程中获取的各属性信息被用于离线学习,离线学习的结果被用于在线转发。此外,为了适应转发策略的需求,传统的决策树算法被修改。3、本文在ndnSIM2.0中部署了上述两种转发策略,并通过与其它策略的对比来评估这两种策略的性能。仿真结果表明:1、基于MDP的转发策略在延迟性能方面优于BEST-ROUTE策略和RANDOM策略,并区分了不同流的重要性;2、通过与OMP-IF等六种转发策略的对比,基于决策树的转发策略在延迟性等方面拥有很好的性能。
[Abstract]:With the emergence of new applications in the Internet and the increasing demand for network services, the traditional TCP/IP-based Internet architecture shows a variety of inadaptability in mobility, scalability and other aspects, especially in the current growing number of applications such as digital media as the emerging distribution service model. To overcome the inadaptability between the current Internet architecture and new applications, researchers at home and abroad have proposed the idea of redesigning the future Internet architecture. As an important member of the future Internet architecture, Named Data Networking (NDN) has a complete routing protocol framework and hierarchical structure. Nomenclature rules have attracted much attention. In NDN, network nodes not only provide forwarding capabilities, but also storage capabilities. The addition of caches in routing nodes not only avoids duplicate data transmission in the network, but also adds intrinsic multipath characteristics to data transmission. The main work of this paper is as follows: 1. Aiming at the disorder problem in the current NDN forwarding strategy, a forwarding strategy based on MDP (Markov Decision Processes) is proposed. In this strategy, the Markov decision process model of data forwarding decision process in a single node is proposed. In the MDP modeling process, the number of different classes of flows in each port is set as the system state to realize the joint perception of network content distribution and port state. 2. Aiming at the deficiency of common perception of network ports and content distribution in the current NDN forwarding strategy, Ben In this paper, a forwarding strategy based on decision tree is proposed, which divides the whole forwarding process into two parts: offline learning and online forwarding. The attribute information obtained in the online forwarding process is used for offline learning and the result of offline learning is used for online forwarding. The results of simulation show that: 1. MDP-based forwarding strategy outperforms BEST-ROUTE strategy and RANDOM strategy in delay performance, and distinguishes the importance of different flows; Compared with the six forwarding strategies such as IF, the forwarding strategy based on decision tree has good performance in delay and so on.
【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.02

【相似文献】