Fine-grained Traffic Measurement For 5G Networks

With the construction and commercialization of 5G networks in recent years,it has supported emerging mobile services such as driverless cars,industrial Internet,HD live streaming and virtual reality,bringing profound changes to many industry sectors including manufacturing,healthcare,education and transportation.However,current 5G networks still need the deployment of specialized hardware devices and software modules from equipment manufacturers in order to meet the varied network quality of service requirements of various industries.This results in high network construction costs and poor customizability.As a result,white-boxing and open source are becoming the development trends of future 5G networks in new vertical industrial applications.On the other hand,because of the high requirements of services for network quality of service in 5G/6G networks,the control plane needs to obtain accurate fine-grained network traffic information in real time to achieve accurate network intelligence control in order to better guarantee the differentiated service requirements of various services.Therefore,this thesis focuses on open source 5G network scenarios to study the fine-grained traffic measurement problem,and its goal is to provide real-time,accurate and fine-grained network traffic information for 5G network intelligent control.Open source 5G networks implement network elements and protocol frameworks using software on a standardized computing platform.In open source 5G networks,network elements are ”programmable”,that is,the functionality of the elements can be changed or added by modifying the source code,which makes it possible to implement accurate,realtime and fine-grained network traffic measurements in the network.However,due to the vast number and high speed of flows in 5G networks,existing single-node measurement solutions are limited by the resources and capabilities on individual network elements to accurately measure large amounts of network traffic.Therefore,Chapter 3 of this thesis proposes an edge node assisted traffic measurement method(TM-EALFI)based on the characteristics of the current 5G open source network architecture,which uses the computational and storage capabilities of the network edge nodes to identify and mark packets of large flows,and then applies different measurement processes to packets of large and small flows at other measurement nodes through which the flows pass to improve the measurement accuracy.Experimental results show that the average relative error of TMEALFI measurements is 30% lower than that of existing classical methods under the same memory conditions.It is very difficult to rely on the resources and capabilities of a single measurement node to measure the flow state of the entire network.Relying only on measurement nodes to independently measure flows passing through their respective streams can in turn lead to the phenomenon of duplicate measurements of flows,resulting in a waste of measurement resources.In contrast,a network-wide measurement task allocation method can coordinate measurement nodes in the network,allowing different measurement nodes to measure different flows and avoiding duplicate measurements.The existing network-wide measurement task allocation scheme do not sufficiently consider the resource constraints of measurement nodes in multiple dimensions and do not consider the scenario of deploying multiple measurement tasks at the same time.Based on the above problems,the measurement task allocation scheme under different number of tasks and different number of resource types is studied in Chapter 4 of this thesis.Firstly,a method ST-MF based on the maximum flow algorithm is designed to solve the measurement task allocation problem in the single-task case.Then a method MTSR-MF is designed for approximating the multi-task measurement solution under single resource constraint based on the ST-MF method combined with the greedy idea.finally,a mixed-task based method MTMR-Mix is designed to approximate the multi-task measurement solution under multi-resource constraint.Experiments show that the ST-MF method is able to compute the same measurement task allocation scheme as the optimal solution,while the difference between the measurement task allocation scheme computed by MTSR-MF and MTMR-Mix and the optimal solution is less than 5%.