Research On Orchestration Of Service Function Chain In Space-Air-Ground Network

With the gradual completion of large-scale commercial deployment of 5G networks,relevant research institutions and commercial companies have begun to conduct research on the nextgeneration mobile communication network 6G.In order to meet the needs of emerging network technologies such as the Internet of Things(Io T)and the Internet of Vehicles(Io V),as well as human expectations for a ubiquitous information environment,Internet of Everything(Io E)has become one of the visions of the next-generation communication network.Therefore,Space-Air-Ground Integrated Network(SAGIN),as an emerging network architecture,has attracted widespread attention.SAGIN is committed to achieving seamless global network connections via integrating space-based networks,air-based networks and ground-based networks.Due to the heterogeneity and time-varying characteristics caused by the initial design of each network segment to operate separately,the resource allocation and service orchestration in SAGIN face many challenges.The emergence of Software-defined Network(SDN)and Network Function Virtualization(NFV)bring potential for better management of the heterogeneous networks.SDN enables centralized management of network devices through a global controller by separating control plane and forward plane,while NFV uses Commercial-Off-The-Shelf to replace traditional dedicated hardware,which greatly improves the flexibility and scalability of network function deployment.Beyond that,Service Function Chain(SFC)is a set of network functions in a specified and predefined order that satisfies specific network services SFC based on SDN/NFV is able to help SAGIN to achieve efficient resource allocation and flexible service provision.With the aid of the above-mentioned emerging technologies,this thesis conducts the following research on the issue of SFC orchestration in SAGIN.First,this thesis studies the placement of SFC based on Deep Reinforcement Learning(DRL)in SAGIN.The SAGIN system model and the SFC model are well designed and a long-term revenue target optimization function constrained by network resources and service requests is constructed.In the models above,the process of placement of the SFC is formulated as a Markov decision process(MDP)and the algorithm for SFC placement based on deep reinforcement learning is proposed to achieve the goal of improving long term revenue through the learning and interaction of agents with the modeled SAGIN.The simulation results show that the proposed algorithm has better performance than the traditional algorithm in terms of service acceptance rate and revenue to cost ratio.Second,this thesis studies SFC placement assisted by federated learning in SAGIN.Aiming at the need to efficiently coordinate physical resources and realize flexible service provision in largescale heterogeneous networks of SAGIN,this thesis proposes to use federated learning to conduct distributed training.By dividing SAGIN into several domains where the local models are trained by deep reinforcement learning.The simulation results show that the service acceptance rate and longterm average revenue can be improved by reducing service delays..