Research On UPF-Based Quality Of Service Guarantee In Mobile Edge Computing

In order to meet the low-latency requirements of emerging applications,the computing resources of edge computing need to be deployed as close as possible to user equipment.As a result,computing resources are distributed geographically at different places.However,the limited resources of a single edge cloud make it impossible to independently respond to all service requirements.Therefore,edge computing systems rely on the coordination across edge clouds to operate,i.e.,improving the utilization of computing resources by efficiently matching the computing requirements on the user side with the computing resources on the server side.In this process,the 5G User Plane Function(UPF),which is the only network function that serve as the uplink classifier in service offloading and the connecting point between the communication network and edge cloud,plays a key role in realizing computing and networking coordination and guaranteeing service quality.Therefore,how to achieve efficient coordination of computing resources and network resources based on UPF and ensure the quality of service in mobile edge computing has become a current research hot topic.Most of the existing work studies the collaborative management of computing resources and network resources in edge computing from the perspective of traditional computer networks,ignoring the differences between the communication network user plane and computer networks in terms of traffic carrying,service offloading,and session management.Besides,since the control planes of the communication network and edge cloud belong to different network domains,it is also challenging to achieve cross-domain coordination.To deal with these challenges,this thesis studies the collaborative management of the communication network and the edge cloud in three levels.Firstly,how to place edge servers and UPFs collaboratively is studied from the perspective of hardware layer.Secondly,the joint management of user request dispatching and container scheduling is studied from the perspective of network layer.Finally,how to reduce downtime of service migration is studied from the perspective of application layer.Main contributions of this thesis are listed as follows:(1)For the joint placement of edge servers and UPFs in 5G mobile edge computing,this thesis studies how to properly place edge servers to cope with the uneven distribution of user requests.The edge server placement problem can be divided into two cases according to whether to add new edge cloud room.In the first scenario,each edge cloud only has one edge server so that it shares cabinet with base station equipment to prevent extra construction cost.This thesis leverages the service level agreement function to develop a profit model that comprehensively considers delay and energy consumption.Then,a particle swarm optimization-based profit-aware edge server placement algorithm is proposed to choose edge server placement locations and assign base stations properly.In the second scenario where extra server room is constructed,this thesis first prunes the solution space and proposes a joint placement algorithm for edge cloud and UPF to reduce the average access delay.Finally,performance evaluations are conducted using real world dataset.The results show that the proposed algorithms can reduce access delay and improve resource utilization effectively.(2)The joint management of request dispatching and container scheduling is studied to dynamically match computing resource provision with users’service requirements.First,this thesis builds a cost model including delay cost,container operation cost,and container switching cost.Then,the request dispatching and service deployment problem is mapped to a Markov decision process,which makes decision on session establishment and container management simultaneously.A request dispatching and container scheduling method is proposed based on deep reinforcement learning.Finally,the performance is evaluated using real word dataset and 5G user plane testbed.The results show that the proposed method can reduce the average cost per slot per request by 4.51%while guaranteeing the delay satisfaction rate.(3)In order to guarantee quality of service during a user moving across different edge servers,this thesis studies how to reduce the downtime brought by service migration.First,the characteristics of applications that may have service migration requirements are analyzed.Second,a measurement is conducted to understand how virtual machine-based and container-based service migration methods work and why these methods may result in unbearable downtime.Third,based on the analysis conclusions of the measurement work,a seamless service migration method based on cloud-edge collaboration is proposed.Finally,an experiment is conducted using cloud game migration to evaluate the performance.The results show that the proposed method can significantly reduce downtime and keep it between 8ms and 14ms.