Research And Implementation Of OpenFlow Network Monitoring System Based On Sampling

The expansion in network size has led to the high complexity of network protocols and management,which has exposed a series of issues in traditional network architecture.In this context,Software-Defined Networking(SDN)dramatically simplifies the management of network by decoupling the network control plane from the network device.In the meantime,SDN network measurement has attracted considerable attention of researchers.OpenFlow protocol supports a standardized southbound interface for direct interaction between the control layer and the data layer,so that the controller can poll the switches to get the flow statistics,but it will also induce significant monitoring overhead on the control channel.Sampling can mitigate the monitoring overhead on the network to achieve effective monitoring of network traffic and perceive the network status in real time.Therefore,how to utilize the sampling for monitoring in OpenFlow Network is the focus of the current research.In this thesis,we focus on the above objectives and implement an OpenFlow network monitoring system based on sampling.The main contents of our contribution are as follows:1.Adaptive Polling Frequency Tuning Based on Sampling:One of the research goals for network measurement is to reduce the monitoring overhead while obtain accurate measurement results.Currently,the controller has to poll the switches periodically to obtain statistics.A high frequency of polling will improve the monitoring accuracy,but it will also induce significant overhead on the switches.In this thesis,we study the tradeoff between measurement accuracy and network overhead,and propose a sampling based adaptive polling frequency tuning method.This method utilizes the sampling action extended from OpenFlow protocol to sample these packets that match the flow entry and estimate its actual traffic.According to the traffic change rate,this approach will gradually adjust the polling frequency.Experimental results show that this method can provide measurement tasks with accurate statistics without incurring too much overhead.2.Detection of Heavy Hitter in OpenFlow Network Environment:The phenomena of a few flows carrying most of traffic has been studied by majority of scholars.As the flow entry is able to count the flow statistics,identifying heavy hitters can rely on querying the traffic statistics.However,the TCAM resource is very limited in switches,so the number of flows that it can track is also limited.Therefore,a large number of flows will influence the accuracy of heavy hitter detection.In traditional network,sampling can greatly alleviate the resources usage,and is widely adopted in network measurement.In this thesis,a sampling-based heavy hitter detection method is proposed.This method utilizes sampling to sample each aggregate flow to identify the potential heavy hitters,and statistically verify the results with newly added flow entries.The evaluation shows that the proposed method can effectively identify the heavy hitters without increasing the usage of flow entries.