Design And Implement Of Network OoS Inference System Based On Data-Driven Method

The increasing complexity of network environment not only puts forward higher requirements for network QoS(quality of service)guarantee,but also brings greater challenges to the evaluation of key performance indicators(such as throughput and delay)concerned by network QoS.Usually,network managers need to conduct a lot of queries and artificial analysis on specific network scenarios in real physical networks to obtain the key performance indicators that network QoS concerns.However,with the increasing heterogeneity and complexity of network systems,this kind of artificial analysis becomes more and more difficult.Traditional network performance evaluation methods include mathematical modeling,network simulation and network simulation,but they can not meet the performance requirements in terms of accuracy and timeliness.At the same time,the rapid development of machine learning,artificial intelligence and other research fields provides new ideas for network performance evaluation.In this context,on the one hand,network managers need an efficient and reliable network performance evaluation method to complete the performance evaluation of specific network scenarios,on the other hand,they want to complete the network performance modeling through machine learning.Therefore,this thesis designs and implements a network QoS inference system based on data-driven.Firstly,investigate the network QoS and its key performance indicators,and using the discrete event driven network simulator OMNeT++to complete the construction of network simulation data set;secondly,investigate the popular technical solutions of network performance modeling through machine learning method in recent years,the graph neural network is selected to be introduced into the field of network delay evaluation,and the network simulation is completed on the basis of network simulation data.Finally,the network delay inference service provided by the network delay performance algorithm model is integrated through the Flask framework to provide network managers with delay prediction in specific network scenarios.On the one hand,the system provides a network delay performance evaluation method based on machine learning method;on the other hand,the system provides a complete network performance algorithm model development process from simulation data set construction to network performance algorithm model modeling,and then to model application and management,which provides a reference for network managers who have custom network performance model development needs for business support.The results of experiment show that the network QoS inference system based on data-driven in this thesis can effectively improve the progress of network performance algorithm development,and can assist network managers to select efficient and highly trusted network performance evaluation model for network performance prediction in specific scenarios.