Research On Low Overhead Wan Measurement Mechanism Based On Software Definition

WAN(Wide Area Network)is an important remote network that connects cross-domain networks and carries a large number of important service traffic for common users and enterprises.How to guarantee the service quality of WAN and monitor the status of network in time is always the focus of research in the industry.In recent years,SDN(Software Defined Networking)has been widely introduced into WAN to meet the requirements of agile operation and dynamic orchestration of cloud service applications,which greatly improves the network service level.At the same time,SDN architecture also brings a new perspective for monitoring solutions.However,most of the current measurement mechanisms based on SDN are designed for networks with single index and regular topology,and they have problems such as incomplete collection of indexes,high cost,low accuracy and poor scalability when applied to WAN.INT(In-band Network Telemetry)is a new monitoring mechanism proposed after the advent of programmable chip.It can realize fine-grained measurement into deep network,but it has disadvantages such as high cost of measurement and limited accuracy of service flows.Therefore,this thesis improves the standard INT and designs monitoring schemes for Underlay network and Overlay network respectively:(1)Comprehensive distributed measurement mechanism based on Underlay network:PingINT is designed by combining Ping with INT,and the measurement initiation and measurement mode switch are realized by deploying proxy on programmable switch.When the network delay status is good,the proxy periodically initiates coarse-grained Ping measurement to the adjacent node to obtain the link delay data of the whole network.In order to determine the congestion contention and other abnormal events in the network more efficiently,this thesis introduces Plateau algorithm to continuously monitor the delay data.Once an exception is identified in the link,the agent immediately sends out multiple INT probe packets to collect indicators such as queue occupancy in different priority queues.In addition,in order to further reduce the cost of data analysis and processing,a fast and slow feedback mechanism is designed according to the degree of data urgency.Finally,the simulation verifies that PingINT can obtain sufficient detailed network information when abnormal events occur,and compared with PingMesh,it can reduce the measurement data cost by 84.16%on average,as well as the control plane command number by five orders of magnitude.(2)On-demand optimal telemetry flow measurement mechanism based on Overlay Network:Due to different Overlay network business flows have different sensitivities to network indicators,network managers are faced with the difficulty of having different measurement requirements in different paths.To solve this problem,this thesis proposes INTLess,aiming to cover the whole network link measurement requirements with the minimum number of telemetry flows and the minimum telemetry data overhead.In the implementation,this thesis adopts segment routing(SR)mechanism and designs "SR+INT" double stack detection probe to realize directional telemetry and customized collection of various network indicators.In addition,INTLess uses greedy algorithm to analyze the whole network measurement demands and plans a group of telemetry flow to meet monitoring demands.In order to avoid greedy algorithm falling into local optimal solution,simulated annealing algorithm is introduced to iterate greedy algorithm several times until convergence to a set of global optimal detection flow.Finally,the thesis verifies through simulation that our telemetry flow planning algorithm can reduce the detection cost up to 53.1%when there are multiple traffic flows in the network.