Research On The Key Technologies Of Service Function Chain Deployment In Network Function Virtualization Environment

With the rapid development and wide use of Internet,innovative communication modes,variational application environments and increasing security requirements propose greater demands on the rapid and flexible deployment of all kinds of network functions.Network function virtualization(NFV)decouples network functions and physical hardware.All kinds of network functions are realized,operated,maintained and managed as virtual network functions(VNF)on universal hardware platform,which overcomes the weaknesses of dedicated function devices,such as functional singleness,high construction cost,difficult deployment and complicated management.NFV has got extensive attention from industry and academia.In NFV environment,researchers arrange virtual network functions by special orders as Service function chains(SFC)for different kinds of business requirement.By the centralized control,resource sharing and on-demand deployment of SFC,diversified and customized services are provided to users.SFC deployment usually consists of two aspects.One is the techonology of function and link mapping for known service requests.The functions and links in SFC are matched with service nodes and transmission paths in physical network,to improve the carrying capacity of network resources.The other is the techonology of service path selection for real-time service requests.The data flows are guided to go through service nodes which can support the functions in SFC in proper order,to optimize the response quality of service requests.In recent years,researchers have studied and explored the problem from the perspectives of saving network resources,optimizing load balancing,reducing end-to-end delay and improving function reliability.However,the algorithms still have some problems in applications,including:(1)The function mapping algorithm is restricted by network resource,so that it is not easy to optimize the resources of nodes and links jointly in the whole network.(2)The computational complexity of link mapping algorithm is too high to deal with a lot of service requests in short time.(3)The service path generation algorithm leads to longer response time of service requests,and reduces the robustness and scalability of network.(4)The end-to-end delay,the reliability of function instances and the load of service node restrict the regulation of service path.To address such challenges,this dissertation studies key technologies of the service function chain deployment in network function virtualization environment.We focus on both the function and link mapping and the service path selection.We try to improve the carrying capacity of network and the response quality of service requests by designing flexible and efficient mapping algorithm of functions and links,and selection mechanism of service paths.The main contributions of this dissertation are summarized as follows:1.Aiming at the problem of restricted node and link resources during the mapping of virtual functions,we propose an optimization algorithm for virtual function mapping based on the longest effective function sequence,which realizes the joint optimization of function and bandwidth utilization.Firstly,with the objective of optimizing function reuse and bandwidth requirement jointly,the problem of virtual function mapping is modeled as an optimal problem under multi-restriction.Then,we modify the longest common subsequence and propose the longest effective function sequence to judge the number of reused functions.Finally,we set the upper bound of path length and search relay nodes incrementally with the most reused functions to place function instances.Simulation results show that,the proposed algorithm can balance network resource and optimize the function and bandwidth utilization rate.Compared with other algorithms,the network resource utilization is reduced by 10%,which can support more service requests.2.Aiming at the problem of high computational complexity of virtual link mapping for a lot of service requests,we propose a virtual link mapping algorithm based on optimal weighted graph matching,which realizes the quick response to a lot of service requests.We combine service function chains and transform the problem of virtual link mapping into the problem of virtual topology matching.We obtain the matching result between graphs of virtual topology and physical topology using eigendecomposition approach,and further optimize the matching result by hill-climbing method.Simulation results show that,the computation complexity is only related to the number of nodes in physical topology.The proposed algorithm can support quick response to service requests,and can balance the load of nodes and bandwidth of links in the whole network.3.Aiming at the problem of long response time of service requests by service path generation algorithm,we propose a service path generation algorithm based on the shortest path,which realizes the real-time processing and forwarding of data flows.Firstly,we give the mathematical description of service path selection problem with the objective of lowest transmission cost.Then,we borrow the idea of the shortest path in traditional routing problem,and design service path routing tables and the process of routing update and lookup.Finally,we theoretically deduce the important parameters in algorithm and analyze the performance and complexity.Simulation results show that,the service path routing tables can generate efficient paths conforming to the service request.The routing item size and the convergence time is linearly related to the number of nodes and the kinds of functions.The cost error between the service paths generated by routing tables and the shortest paths is less than 25 percent,and the response delay to service requests is reduced a lot4.Aiming at the problem of service path regulation problem with the restriction of multi parameters,we propose a service path regulation algorithm based on reinforcement learning,which realizes the optimal solution and efficient regulation of service paths.Firstly,the problem of service path with multi parameters is mathematically modeled as a combinatorial optimization problem.Then,we design a reinforcement learning model for service path,and solve the problem of optimal service path by designing action selection policy,reward function and quality function.Finally,we design the service path regulation mechanism according to the optimal service paths calculated by reinforcement learning algorithm.The default paths are generated by service path routing tables.As soon as a path needs modification,controller sends regulation messages to nodes on path to modify their forwarding objectives.Simulation results show that,the reinforcement learning model can optimize the delay of path,reliability of SFC and load balancing jointly.On the basis of service path routing tables,the regulation mechanism can guarantee the service performance,such as response delay and cost of path,and load balancing of nodes in network.