SDN数据面调度机制性能建模与用户体验质量优化的问题研究

The provisioning of quality of experience in Cloud Centers is a thrust of changing network landscape. Software defined networking, SDN, has emerged as a promising network technology which enables the dynamic nature of future network functions and intelligent applications while lowering operating costs through simplified hardware, software and management. Due to the advantage of SDN and the increasing demand, researchers have made a lot of efforts, such as the studies on OpenFlow protocols and related works, ForCES, controller design, forwarding devices, southbound and northbound APIs, the emulation and simulation tools, and of course the applications. However, it is barely to find research works on the performance of scheduling mechanisms which regulate the forwarding of traffic flows in data plane. Understanding the performance and limitations of the forwarding capability in data plane is a prerequisite for using it effectively. More importantly, the performance analysis of resource allocating policy is an enabler for evaluating quality of experience. ..To this end, we will investigate the performance of various scheduling mechanisms in SDN data plane. Further the developed models will be applied on evaluating and optimizing quality of experience in SDN. The feasibility and accuracy of scheduling models replies on how the traffic and forwarding capability are characterized. Unfortunately, we can hardly find an effective traffic model which closely characterizes the SDN traffic. In addition, there are few works reported for modeling the variability of SDN switch forwarding capability in data plane. To fill the gap, this work will focus on the following aspects: 1. Modeling the characterizations of SDN traffic flows and its variable flow length; 2. Modeling the variable forwarding capability of SDN switches, and investigating how its variable feature affects the network performance and quality of experience; 3. Modeling scheduling mechanism with the above traffic and forwarding capability models, and apply analytical model on evaluating the network performance and optimizing the quality of experience. ..It can be readily seen that this work is a frontier study of the corresponding area. It fills the gap of research by taking into account the above three issues. This work will provide not only the theoretical basis, but also a tool for designing software defined networking, which will fill the impressing demand.

在软件定义网络数据面中，合理调度数据流以及数据和控制面间的交互流量对整个软件定义网络的设计、性能和用户体验质量起着关键的作用。然而文献中对这个问题的研究存在着问题：1. 对软件定义网络数据面的流量分析仍借用传统网络流量模型；2.缺少数据面节点转发容量变化特性对网络性能和用户体验质量影响的研究；3.缺乏对结合了软件定义网络数据流模型和可变的转发容量的调度机制的性能研究，以及利用模型来优化网络性能和用户体验质量。本项目针对解决上述三个问题进行研究，其重要意义在于：1.在软件定义网络中，目前尚无结合以上三个问题的研究；2. 有重要的理论意义和重大的应用价值，性能建模对软件定义网络的设计有指导作用，能评估和优化用户体验质量；3. 具有时间紧迫性，软件定义网络的构架已经广泛出现，但仍然缺乏对其网络性能的评价理论和工具，因此对软件定义网络数据面调度机制建模和用户体验质量评估的需求日益迫切。

软件定义网络数据面的性能分析考虑用户流量对单一转发节点的服务性能的影响。此外，鉴于软件定义网络中数据流的特性，用户数据流在到达目标前，可能会经过多个中途路由转发节点。本项目评估和预测流量对各个中途路由节点服务性能的影响，建模了流量在中途节点的叠加和剥离。并将流量叠加和剥离方法应用于城市路网交通流量的建模中，取得了比较好的效果。在建模转发节点的服务能力方面，本项目采用指数分布随机过程。实验验证结果表明，指数分布随机过程很好的刻画了转发服务能力的波动性和随机性。基于以上两点，本项目应用改进排队论，以用户到达流量与服务能力为参数建立单一服务、单一到达队列的系统。解析上述的队列系统可以获得队列长度分布和等待时间分布这两个重要的性能指标，从而评价该节点对于到达流量的服务质量。更进一步，项目的值得指出的是，模型中使用的流量模型考虑到了自相似和长相关特性，自相似特性是被广泛证明存在网络流量的显著特征。本项目的另一个重要研究结果是基于流量模型和排队论评估数据面和控制面之间的交互流量对转发节点的性能影响。本项目讨论和实现了Jackson模型在调度数据面和控制面板交互流量的性能，并进一步提出以优先级调度来实现数据面与控制面板交互流量，结果证明，本项目提出的调度方法性能相对Jackson模型有优势，尤其是在数据流量存活期有严格限制的情况下。建模优先级调度的难点在于解决多个队列共享转发资源的耦合问题，本项目提出一种基于空队列估计的解耦方法，将耦合负责系统分解成一组单一队列系统，从而获取每个队列的性能指标。项目的完成为软件定义网络数据面用户体验质量与转发能力的性能评价提供理论支撑和技术方法。