Research On Key Technologies For Cloudified Mobile Communication Networks

As diverse,differentiated,and customized service demands from vertical industries continue to thrive,mobile communication networks(MCNs)will not only serve as data transmission channels but also innovation-driven service platforms with flexibility,agility,and openness.To meet these visionary needs,the concept of cloud-network convergence has been proposed,which essentially achieves the integration of cloud computing,MCNs,big data,and artificial intelligence(AI)to assist vertical industries in digital transformation.To fully leverage the advantages of cloud computing resources and technology,the architecture of MCNs needs to be decoupled and reconstructed towards cloud computing to form a convergent,service-oriented architecture that supports lightweight,cloud-native deployment.At the same time,to ensure the robustness of cloud-native MCNs under naturally fluctuating cloud computing resource scenarios,research on intrinsically intelligent network re-configuration strategies is required to form an online closed-loop intelligent,automated control process.Therefore,the cloud-native deployment and robustness guarantee of MCNs have become urgent problems to be solved.To achieve stable and efficient cloud-native deployment of MCNs,this thesis theoretically analyzes the impact of cloud computing lightness,distribution,and dynamism on cloud-native MCNs,and investigates three key technologies,namely,converged service-based architecture(cSBA),intelligent decision mechanism for random network delays,and intelligent re-configuration strategies for cloud-native MCNs,respectively.To verify the feasibility of the proposed methods,mechanisms,and strategies in real networks,an open-source system for cloudified mobile core networks is developed.The main research work and innovative results are as follows:1.Converged Service-Based Architecture for MCNsTo address the issue of lightweight cloud-native deployment of MCNs,this thesis proposes an method for decouping network functions from the perspective of the protocol stack that is compatible with the layered structure of the terminal air interface protocol stack and redundant functions between the upper layer protocols of the radio access network(RAN)and the 5G core network,and designs the cSBA.The simulation results show that 1)the architecture can save about 63.5%of CPU resources compared with the monolithic architecture under periodic small traffic scenarios with almost the same processing performance;2)the architecture can improve the performance about 2.5 times compared with the monolithic architecture under continuous growth traffic scenarios with the same resource usage.Based on the architectural,this thesis proposes the service-based N2 mechanism and its enabled message-level stateless 5G core network.Experimental results show that this mechanism enables the cloud-native deployment of 5G core network with the ability to cope with service instance anomalies and the dynamic nature of cloud computing resources,which initially verifies the feasibility of this architectural theory.2.Intelligent Decision Mechanism for Random Network DelaysTo address the issue of AI-native management of distributed cloud services,this thesis abstracts the online closed-loop control process with random interaction delays and models it as a partially history-dependent Markovian decision process,designs random-delay-corrected deep reinforcement learning(DRL)algorithm framework and proposes an intelligent decision mechanism for random network delays.Based on the DRL principle of this mechanism,this thesis investigates the pipeline task offloading strategy in Cybertwin-enabled networks.This thesis models the problem as a standard Markov model and proposes the random-delay-corrected multi-agent DRL algorithm for optimal solution through model transformation.Simulation results show that the proposed algorithm has significant performance advantages over the DRL algorithms without considering the random delays in terms of optimal convergence,decision reaction time,task offloading success rate and average task completion system consumption for different transmission delays,different number of terminals and different number of edge servers.3.Dynamic Adaptation Strategies for Cloud-Native MCNsTo address the impact of cloud computing resource dynamics on the robustness of cloud-native MCNs,this thesis proposes a unified model where traditional cloud services and MCN functional services are deployed together in public cloud platform and proposes the dynamic adaptation strategies for cloud-native MCNs,including adaptive scaling of network resources and adaptive scheduling for signaling traffic that is based on the message-level stateless mechanism.In terms of resource scaling,this thesis models the problem as a cumulative quality-of-service optimization problem with constraints,where network resource sensitivity metrics are defined,and proposes an adaptive resource scaling strategy based on multi-pass deep Q networks(MP-DQN).Simulation results show that this strategy can not only guarantee the success rate of cloud task offloading but also reduce the failure rate of services compared with methods that do not consider sensitivity metrics,thus improving cloudnative MCNs’ stability.In terms of traffic scheduling,this thesis models the problem as a joint optimization problem of signaling traffic throughput and average service time with constraints and proposes an adaptive traffic scheduling strategy based on the proximal policy optimization(PPO)algorithm,in which a graph attention mechanism(GAT)is used to extract the complex environment state.The simulation results show that this strategy has a greater performance advantage than the scheduling strategy based on the procedure-level stateless mechanism in terms of signaling procedures normal processing rate and average completion time,and it is more adaptable to the dynamic nature of the cloud environment because it can maintain stable signaling processing performance under the presence of uncertain anomalies in the system.4.Open-Source Prototype for Cloudified Mobile Core NetworksTo validate the above three key technologies,this thesis develops an opensource cloudified mobile core networks.For the agile software delivery and capability update brought by cloud computing,this thesis designs and builds a a code development and Operation(DevOps)platform based on Gitlab and Jenkins.For lightweight cloud-native deployment,this thesis designs the software architecture of core network functions based on high-performance InTerTaskInterface(ITTI)and Pistache;for core network protocol stacks,this thesis implements the basic registration,session establishment,bearer establishment,and service data transmission procedures according to the standard;for highperformance data-plane transmission problems,this thesis designs and implements the data-plane functions based on vector packet processing(VPP);for service-based N2 mechanism,this thesis designs and implements a middleware with message rate awareness(MRA)and message feedback confirmation(MFC)mechanisms to realize the message-level stateless 5G core.To validate the intelligent decision-making mechanisms and network re-configuration strategies,this thesis 1)customizes and modifies the Kubernetes-based cloud environment to realize the online closed-loop control process for periodic collection of environmental states and issuance of network control commands;2)designs an AI engine(AIE)that can automatically load DRL algorithms and apply different policy algorithms;3)uses Cpulimit and ChaosBlade to simulate the dynamic nature of cloud computing resources and random network delays.Finally,the feasibility of these technologies are verified by an adaptive resource scaling use case.