无线数据中心网络架构及其调度机制研究

发布时间：2018-07-31 05:54

【摘要】：数据中心网络是近年来的新兴的云计算中提供业务支持的重要基础平台。当前主流的基于以太网的数据中心网络的性能严重受制于由不均衡的流量分布造成的网络拥塞问题。由于拓扑结构是固定的，对于产生高额流量的热点节点而言，它们并不能获得与需求所匹配的可用带宽，而由此产生的网络延迟会进一步影响整个系统的性能。与以太网不同，，无线网络的拓扑结构可以灵活的进行调整，因而可以利用这个特征动态的满足热点的流量需求。本文围绕在数据中心中引入无线网络以优化其性能的问题展开工作，设计了支持无线传输的数据中心网络架构，研究了其中最为核心的无线调度问题。本文探讨了在数据中心中使用无线的可行性，设计了一套有线/无线混合的数据中心网络架构EWHA及相应的分布式的无线调度机制，通过以太网与无线网络的协同工作，高效的根据流量分布的变化改变无线网络配置，解决热点的高流量引起的带宽供给不足问题。针对调度机制中最为关键的无线链路的建立和信道分配问题，本文基于避免冲突无线传输的协议干扰模型，对无线数据中心网络进行建模，并综合考虑了天线、信道以及干扰等多方面因素；提出了最小化热点流量和最大化网络效用两种解决思路，分别提供了一个贪心算法MMS和一个基于匹配理论的迭代算法MUS进行求解。模拟实验结果表明本文的解决方案能够显著的利用无线传输提高数据中心网络整体性能。本文将协议干扰模型的方案扩展到更为实际的、考虑到不同干扰强度的物理干扰模型中，通过引入信干噪比以及可变无线传输速率等因素进行扩展，进一步深化了基于最大化网络效用调度无线的方案，提出了一个优化设计的遗传算法GAS和基于松弛-对齐理论的解决方案RRS，前者能够高效的处理随时间进化的流量分布，而后者能够获得性能不低于最优解1/2的调度方案。通过详细的性能评价，验证了这两个算法能够有效的解决无线调度问题，同时改善网络的吞吐量等性能指标。
[Abstract]:Data center network is an important basic platform to provide business support in cloud computing in recent years. The performance of the current mainstream ethernet-based data center networks is severely constrained by network congestion caused by uneven traffic distribution. Because the topology is fixed, they can not get the available bandwidth to match the requirements for hot nodes which generate high traffic, and the resulting network delay will further affect the performance of the whole system. Different from Ethernet, the topology of wireless network can be adjusted flexibly, so it can be used to dynamically meet the traffic requirements of hot spots. This paper focuses on the problem of introducing wireless network into the data center to optimize its performance, designs the data center network architecture to support wireless transmission, and studies the most core wireless scheduling problem. This paper discusses the feasibility of using wireless in the data center, designs a wired / wireless hybrid data center network architecture EWHA and the corresponding distributed wireless scheduling mechanism, and works together through Ethernet and wireless network. Efficiently change the wireless network configuration according to the change of traffic distribution, solve the problem of bandwidth shortage caused by hot high traffic. Aiming at the establishment of wireless link and channel assignment which is the most critical problem in scheduling mechanism, this paper models the wireless data center network based on the protocol interference model to avoid collision, and considers the antenna synthetically. In this paper, we propose two solutions, including channel and interference, to minimize hot spot traffic and maximize network utility. A greedy algorithm (MMS) and an iterative algorithm (MUS) based on matching theory are proposed to solve the problem. Simulation results show that the proposed solution can significantly improve the overall performance of data center network by wireless transmission. In this paper, the scheme of protocol interference model is extended to be more practical, and considering the physical interference model with different interference intensity, the signal-to-noise ratio and variable wireless transmission rate are introduced to expand. This paper further deepens the wireless scheme based on maximizing network utility scheduling, and proposes a genetic algorithm (GAS) for optimal design and a solution based on loose-alignment theory, which can efficiently handle the traffic distribution over time. The latter can obtain a scheduling scheme with performance not less than 1 / 2 of the optimal solution. Through the detailed performance evaluation, it is verified that these two algorithms can effectively solve the wireless scheduling problem and improve the network throughput and other performance indicators.
【学位授予单位】：清华大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP308

【相似文献】