数据中心光互联网络中基于OpenFlow的跨层调度机制研究

发布时间：2018-06-18 19:01

本文选题：数据中心 + 光互联　；参考：《北京邮电大学》2014年硕士论文

【摘要】：目前在数据中心(Data Center, DC)中,互联网和新型应用的快速发展,带动海量的数据流量增长,使得数据中心在网络时延、带宽和可靠性等方面提出了更高的要求。传统的电交换网络越来越难以满足新型数据中心的需求,而有着低时延、高带宽特点的光交换网络开始崭露头角,得到广泛应用。在数据中心光互连网络中,集中式控制方式有着较强的灵活性和可扩展性,因此而得到了广泛的亲睐。开源的OpenFlow是实现集中式控制最为成熟的手段。集中式控制和OpenFlow协议相结合,可在数据中心光互联网络中,进行富有想象力的创新。另外,从网络资源调度角度讲,传统网络中各层资源分层控制和调度,使得各层资源利用不够充分。而在基于OpenFlow的集中式控制网络中,实现资源的跨层调度变得简单易实现,跨层调度可以提高网络资源的利用率,大大简化流的处理流程。在数据中心网络中,TCP incast是普遍存在于多对一通信模型中的冲突问题。TCP incast发生后,网络吞吐率会出现严重的崩溃现象。本文针对TCP incast问题提出了一套在数据中心光互联网络中的跨层调度解决方案。利用OpenFlow的集中控制和跨层调度,使得TCP incast吞吐率崩溃的问题得到有效的缓解。首先,本论文基于光突发传送网(OBTN)的数据中心光互连网络,发现网络中TCP incast问题主要存在于节点业务接入侧和OBTN环网内,并分别制定了解决方案。在业务接入侧,本论文设计了基于OpenFlow的跨层调度机制。借助OpenFlow网络控制器和应用层业务控制器协同控制,即通过监控节点缓冲区的占用率,进而控制多对一通信模型中并发服务器的数量,可以有效避免数据中心内光互联网络中TCP incast问题。针对该跨层调度机制,本论文完成了OpenFlow协议拓展,并对跨层调度机制的性能做了仿真验证。其次,在OBTN环网内,结合OpenFlow网络控制器的业务监控,本论文提出了针对incast问题的改进型动态带宽分配(DBA)算法。该算法将所有相关的incast业务作为一个整体,优先为其授权和配置带宽资源,可以有效改善incast业务的应用层性能,并提高整个网络的吞吐率。本论文通过数值仿真和网络仿真验证了改进型DBA算法的性能。
[Abstract]:At present, with the rapid development of the Internet and new applications in data Center (DCD), massive data traffic is increased, which makes the data center require higher network delay, bandwidth and reliability. Traditional switching networks are becoming more and more difficult to meet the needs of new data centers. Optical switching networks with low delay and high bandwidth are emerging and widely used. In the data center optical interconnection network, the centralized control mode has strong flexibility and scalability, so it has been widely used. Open source OpenFlow is the most mature means to realize centralized control. The combination of centralized control and OpenFlow protocol enables imaginative innovation in the data center optical Internet. In addition, from the point of view of network resource scheduling, the hierarchical control and scheduling of each layer of resources in the traditional network make the resource utilization of each layer insufficient. In the centralized control network based on OpenFlow, the realization of cross-layer scheduling of resources becomes simple and easy. Cross-layer scheduling can improve the utilization of network resources and greatly simplify the flow process. In data center networks, TCP / incast is a common collision problem in a many-to-one communication model. After TCP / incast occurs, the network throughput will collapse seriously. In this paper, we propose a cross-layer scheduling solution for incast in data center optical interconnection networks. Using the centralized control of OpenFlow and cross-layer scheduling, the problem of TCP incast throughput collapse can be effectively alleviated. Firstly, based on the data center optical interconnection network of optical burst transport network (OBTNN), we find that the problem of incast in the network mainly exists in the access side of node services and the OBTN ring network, and the solutions are worked out respectively. In the service access side, this paper designs a cross-layer scheduling mechanism based on OpenFlow. With the help of OpenFlow network controller and application layer service controller, the number of concurrent servers in a multi-to-one communication model is controlled by monitoring the occupancy rate of node buffer. It can effectively avoid the problem of incast in the optical network in the data center. In this paper, the OpenFlow protocol is extended and the performance of the cross-layer scheduling mechanism is verified by simulation. Secondly, an improved dynamic bandwidth allocation algorithm for incast problem is proposed in this paper, combined with the traffic monitoring of OpenFlow network controller in OBTN ring network. This algorithm takes all the related incast services as a whole and gives priority to authorize and configure bandwidth resources. It can effectively improve the performance of the application layer of incast services and improve the throughput of the whole network. The performance of the improved DBA algorithm is verified by numerical simulation and network simulation.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP308

【参考文献】