Research On Key Technologies Of Flow Measurement In SDN

Flow measurement is one of the important research branches of network performance management.It can provide real-time basic measurement data for many management applications,such as traffic engineering,abnormal traffic detection and traffic behavior analysis,so as to ensure the reliable operation of these management applications.With the rapid development of information technology,both network environment and network technology are changing with each passing day,such as the increasing prosperous applications has led to a sharp increase in network traffic,the network environment has evolved and gradually extended to the Internet of Things,SDN can greatly improve the efficiency of network management by decoupling control and data plane,and so on.These changes profoundly affect the working mechanism,deployment mode and the specific measurement requirement for upper management applications,which drives the flow measurement technology to change and innovate.Currently,SDN has been widely popularized and applied to a certain extent.This thesis mainly studies the high-efficiency flow measurement technology in SDN.Firstly,it proposes a flexible monitoring framework in SDN.Secondly,it proposes two flow statistics collection methods from the perspective of the whole domain of SDN controller and from the perspective of the single SDN switch respectively.Finally,based on the monitoring framework and the flow statistics collection methods,it proposes a large flow detection method.The main innovations and work of the thesis are summarized as follows:(1)The monitoring framework in SDN.A monitoring framework named FlexMonitor in SDN is proposed,which consists of four parts: the analysis of measurement tasks,the deployment of measurement strategies,the collection of measurement data and the analysis of measurement data.By introducing the event triggering mechanism,Flex Monitor not only enriches the deployment modes of monitoring strategies,but also extends the kinds of monitoring data sources,which greatly enhances the flexibility of the monitoring framework.Based on FlexMonitor,a DDoS attack detection method is proposed to verify the effectiveness of Flex Monitor.The experimental results show that the DDoS attack detection method has good detection performance and indirectly show that FlexMonitor can flexibly support the design of a variety of flow measurement methods.(2)The flow statistics collection from the perspective of the single SDN switch in SDN.Taking full advantage of various monitoring mechanisms provided by OpenFlow,a low-overhead flow statistics collection method named EffiView is proposed from the perspective of the single SDN switch.EffiView mainly relies on the flow statistics triggering mechanism to realize the flow statistics collection.This mechanism can not only capture the sudden change of flow traffic in time,but also reduce the measurement overload generated in the collection process.In addition,as an effective complement to the flow statistics triggering mechanism,EffiView also uses two other mechanisms: the flow statistics request message and FlowRemoved message,in order to ensure the fine-grained flow statistics collection.Compared with the traditional flow statistics collection methods,the experiment results show that this method not only has less deployment overhead,but also generates less measurement load in the collection process while ensuring the accuracy and timeliness.(3)The flow statistics collection from the perspective of the whole domain of the controller in SDN.This paper presents a low-overhead flow statistics collection method named EffiCover from the perspective of the whole domain of the controller.On the one hand,this method proposes a mechanism that the controller can maintain the active flow list of the whole domain in real time avoiding the partial loss of network view caused by the use of mask rules in switches;on the other hand,according to the maintained active flow list of the whole domain,this method proposes a target switches selection algorithm.This algorithm can ensure that the active flow statistics of the whole domain can be collected with low overhead,while avoiding the uneven distribution of the measurement load in the network.The experiment results show that this method can achieve the full-network view with low overhead in SDN using mask rules.(4)The large flow detection in SDN.An efficient elephant flow detection method named EffiEye based on the flow features and the flow statistics threshold is proposed.On the one hand,based on the flow features,the controller can identify some large flows in order to narrow the detection scope of large flows.On the other hand,aiming at the flows that cannot be identified as large flows based on the flow feature analysis,this method proposes a large flow detection algorithm based on two different flow statistics thresholds.The algorithm can avoid detection errors caused by the use of mask rules in SDN switches.Through in-depth theoretical analysis,the large flow detection method has strong feasibility and can achieve accurate and timely large flow detection with low cost.