Research And Implementation Of Key Technologies For Software-defined Network Measurement

With the rapid development of the Internet,network service quality has been increasingly emphasized,and a network measurement system is a software system that provides network service quality measurement.In the field of network measurement,the performance parameters of the network layer are delay,packet loss and their extension indicators,while different applications in the application layer have different measurement methods,measurement indicators and statistical methods.Traditional network measurement systems are difficult to serve the diverse application measurement needs,for example,in WebRTC video applications,the measurement methods of active detection and RTCP protocol feedback are used to measure the delay of a set of packets gradient and the number of consecutive packet drops in a fixed period of time,and in QUIC protocolbased applications,the estimated value of round-trip delay and the data consumption rate at the receiving end are measured using a passive analysis of the feedback UDP packets.It is thus clear that application layer measurements require more flexibility in terms of measurement methods,measurement metrics and statistical methods and other measurement needs compared to network layer measurements.In the scenario of diverse measurements at the application layer,fine-grained measurement requirements generate more active measurement tasks,and mutual interference such as resource competition between active measurement tasks may affect the reliability of the measurement data,while in networkoriented measurements,the measurement tasks are simple and have a single configuration parameter,and better scheduling results can usually be achieved using scheduling such as first-come,first-served and random scheduling.In application-oriented measurements,designing reliable measurement task scheduling algorithms is an important way to solve the problem of measurement tasks interfering with each other.In summary,designing flexible network measurement systems for diverse measurement scenarios has important practical application value for improving the quality of application services.Software-Defined Measurement(SDM)is the use of dynamic configuration and programmable measurement methods to improve the flexibility of network measurement,including the flexibility of measurement methods,measurement metrics and statistical methods.The control plane has the ability to maintain data reliability and system resource management,while the data plane provides general support for basic measurements,which is highly flexible and scalable in network measurements,and is therefore widely used in the design of measurement system architectures for diverse measurement needs.This paper draws on the design ideas of software-defined networks,analyses the diverse application-oriented measurement scenarios,proposes the system architecture and detailed requirements of the software-defined network measurement system,analyses the resource conflict problems between measurement tasks in the measurement system and the specificity of measurement tasks,proposes the network measurement task scheduling model and algorithms for the measurement system in this paper,uses measurement task scheduling and measurement process monitoring techniques to propose measurement The accuracy and completeness of the measurement data are evaluated using measurement task scheduling and measurement process monitoring techniques.The functionality and performance of the measurement system is tested in a real WAN environment.The main work of this thesis is divided into the following areas.(1)To propose and implement a software-defined network measurement system architecture for diverse measurement needs.The system is divided into control,management and execution planes greatly decoupling the control,management and execution processes of network measurement,effectively enhancing the flexibility of network measurement.The control plane is responsible for resolving user requirements and scheduling measurement tasks;the management plane is responsible for managing the topology of measurement tasks,monitoring the measurement process and automatically deploying system measurement components;and the execution plane is responsible for executing measurement tasks and uploading and storing measurement data,which supports the ability to provide diverse measurements and a flexible measurement system architecture for the application layer compared to software-defined measurement technologies such as OpenSketch and TPP..(2)To analyse the characteristics of active network measurement tasks in concurrent scenarios and propose a multi-time-slice network measurement task scheduling model and a single-time-slice task scheduling algorithm with expected time priority,which consumes an additional 14%of scheduling time and improves measurement data accuracy compared to earlier chance-balanced scheduling algorithms 15.27%to 98.28%.(3)Propose a method to assess the accuracy and integrity of measurement data,design measurement process monitoring techniques and measurement environment monitoring techniques to identify abnormal measurement data,design monitoring and alerting techniques for data transmission and storage,and assess the data accuracy and integrity of measurement data.