软件定义网配置一致性研究

发布时间：2018-08-15 12:41

【摘要】：软件定义网络作为一种未来网络架构将控制和数据平面分开,使用集中控制机制可以有效降低网络变动时的网络异常。控制器收集全局网络拓扑和状态信息,根据用户策略和算法生成网络配置,然后安装到转发设备的流表中。交换机根据自身的配置处理数据包。虽然软件定义网络使用集中控制,但是转发设备仍然按照本身流表进行转发,也就是说在数据转发平面上,软件定义网络仍然是一个分布式系统。配置更新过程中的配置不一致性仍然会造成网络的不稳定。即使新(旧)配置的正确性得到保障,复杂多样的临时网络状态仍然会导致路由冲突和传输错误。因此在变动的情况下,一致性需要得到保证。网络配置一致性更新机制虽然已经研究了很长时间并取得了一些成果,但是大部分的研究集中在传统网络中,并且针对特定的协议。对于软件定义网络中的配置一致性研究主要集中在最近几年,按照所用的方法更新机制可以分为:标签、序列化和依赖图更新机制。标签更新机制使用标签技术,对配置添加标签以区别新旧配置,对数据包同时添加标签来匹配不同的配置。序列化更新机制将更新按照一定的顺序安装。依赖图更新机制将资源分配图引入更新机制中,使用图来表示更新之间的,更新与路径之间的,更新与链路之间的关系等等。其中包级别配置一致性方案不适用于增量更新;流级别更新方案使用两阶段更新机制和周期性分割机制,需要的时间太长。本文中,我们针对包级别更新提出了基于关系图的更新机制:首先对更新之间的关系进行分析,然后生成更新之间的关系图,接着尝试从关系图中找到一个合适的更新顺序满足包级别配置一致性。针对流级别更新提出基于K前缀覆盖的更新机制:控制器收集已有的流信息,然后计算这个流集合的最佳K前缀,接着将新配置以低优先级安装并且将K条规则碎片以高优先级安装。为了得到最佳K前缀,我们设计了K前缀覆盖算法。该算法计算具有最小覆盖空间的K前缀集合,这些前缀集合包含所有的流。实验结果表明,我们提出的更新机制加快了配置更新的速度,减少了规则安装的规模以及控制消息的规模。
[Abstract]:As a future network architecture, the software defines the network to separate the control from the data plane, and the centralized control mechanism can effectively reduce the network anomalies when the network changes. The controller collects the global network topology and state information, generates the network configuration according to the user policy and algorithm, and installs it into the stream table of the forwarding device. The switch processes packets according to its own configuration. Although the software defined network uses centralized control, the forwarding device still forwards according to its own stream table, that is, on the data forwarding plane, the software defined network is still a distributed system. Configuration inconsistencies in the configuration update process will still result in network instability. Even if the correctness of the new (old) configuration is guaranteed, complex and diverse temporary network states will still lead to routing conflicts and transmission errors. Therefore, in the case of change, consistency needs to be guaranteed. Although the network configuration consistency updating mechanism has been studied for a long time and some achievements have been made, most of the research is focused on the traditional network and is specific to specific protocols. The research on configuration consistency in software-defined networks is mainly focused in recent years. According to the methods used, the updating mechanisms can be divided into tag, serialization and dependency graph updating mechanisms. Tag updating mechanism uses tag technology to add tags to the configuration to distinguish between new and old configurations and to add tags to packets to match different configurations. The serialization update mechanism installs updates in a certain order. The dependency graph update mechanism introduces the resource allocation graph into the update mechanism and uses the graph to represent the relationship between update and path update and link and so on. The package level configuration consistency scheme is not suitable for incremental update, while the flow level update scheme uses two-stage updating mechanism and periodic partitioning mechanism, which takes too long. In this paper, we propose a graph-based update mechanism for package level updates: first, we analyze the relationship between updates, and then we generate a graph between updates. Then try to find an appropriate update order from the diagram to meet package level configuration consistency. An update mechanism based on K-prefix coverage is proposed for stream level updating. The controller collects the existing stream information and then calculates the best K-prefix of the stream set. The new configuration is then installed at a low priority and the K rule fragment is installed at a high priority. In order to obtain the best K prefix, we design a K prefix covering algorithm. The algorithm computes K-prefix sets with minimal covering space, which contain all streams. The experimental results show that the proposed updating mechanism speeds up the configuration update and reduces the size of the rule installation and the control message.
【学位授予单位】：清华大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP393.02

【相似文献】