Research On Routing And Resource Optimization For Service Function Chain In SDN/NFV-enabled Networks

Recently,the business model of network operators has undergone a revolutionary transformation resulting from the appearance of new network technology and various demands of network users.Network Function Virtualization(NFV)and Software Defined Network(SDN)have been proposed as two viable techniques in this transfor-mation.Through virtualization technology,NFV decouples network functions from dedicated hardware,and allows to run them as software in commercial-off-the-shelf devices,which has significantly enhanced the flexibility and scalability of network service placement.SDN separates control plane from data plane,and it is convenient to achieve centralized management and control for devices in networks.Thus,using SDN and NFV,such new network architecture can flexibly place services and achieve efficient resource allocation and optimization by managing network devices in a centralized way.Service Function Chain(SFC)is a network technique that is being standardized by Internet Engineering Task Force,it needs flows to traverse a series of VNFs in a specific order,and network operator can utilize SFC to provide users with customized network services.However,there still exist many problems and challenges for the usage of SFC.In SDN/NFV-enabled networks,in order to enhance the network performance and achieve load balancing,the same kind of VNF can be flexibly placed in different network devices,which results in the scenario of VNF multiinstance placement.During the routing computation for SFC,how to select the optimal VNF Instances(VNFI)in the scenario of VNF multi-instance placement and how to compute the best path to satisfy the order constraint of SFC have been important problems for academic research.Considering dynamic change of network load,network operators need to self-adaptively adjust and optimize the number and location of VNFIs placed in the network to improve network performance.Moreover,according to the network conditions and SFC routing strategies,great VNF placement approaches can save the resource consumptions in links and nodes by optimizing the number and locations of placed VNFIs,which can further provide good service environment for network applications and data transformation.Nevertheless,the VNF placement problem has been proven to be NPhard.Thus,for complex changes of network conditions,how to design flexible and efficient VNF placement strategies is another challenge we have to address.This dissertation mainly studies the problems of SFC routing,VNF placement and resource optimization in SDN/NFV-enabled networks.Given ditterent considerations of performance,the research contents of this dissertation include the following three parts:the research of SFC routing based on the match of flow characteristics,the research of dynamic VNF placement and routing based on self-adaption of network load and the research of SFC routing and resource optimization based on machine learning.The third research consists of two small research points:the research of SFC routing based on deep learning and the research of VNF placement and resource optimization based on deep reinforcement learning.The work and major contributions of this dissertation are listed below:1)For the characteristics of resource preferences of network flows,this research defines the relative cost and proposes a routing algorithm for SFC based on the match of flow characteristics to enhance network performance and achieve the balance of network load.In order to achieve finegrained routing computation for the flows of SFC services,this research analyzes the characteristics of flows and classifies them according to bandwidth and computation requirements.And the concept of relative cost is defined according to network resource conditions.Then,we take full advantage of central management of SDN and establish a mathematical model for the optimization of rout-ing problem to balance network load.Further,the Resource Aware Routing Algorithm(RA-RA)is proposed to provide differentiated routing services for the flows of SFC services.Performance evaluation shows that RA-RA can get near optimal performance in small-scale network and it runs much faster than the optimal algorithm.In addition,RA-RA can improve network performance obviously than compared algorithms.2)For the characteristics of dynamic change of network load and the flexible placement of VNF,we jointly consider the SFC routing and VNF placement problem and establish a mathematical model for it.Then,utilizing the relative cost,the SFC MAPping(SFC-MAP)algorithm and VNF Dynamic Release Algorithm(VNF-DRA)are proposed which minimize the sum cost of the objective and enhance the utilization ratios of network resources.Considering the influences of multiple kinds of resources,dynamic network load and load balancing,this research studies the dynamic VNF placement and routing problem.We establish a mathematical model for this problem whose aim is to minimize the VNF placement cost and resource cost during the routing computation of an SFCR.Further,according to the idea of graph transformation,SFC-MAP is proposed.By the usage of multi-layer graph,SFC-MAP can jointly compute the paths for SFC and achieve dynamic placement of VNFIs.Moreover,VNF-DRA is proposed to periodically release redundant placed VNFIs according to the change of network conditions.Performance evaluation shows that,SFC-MAP&VNF-DRA can obtain near optimal results in small-scale network and its execution speed is faster than the optimal algorithm.Moreover,compared with other algorithms,SFC-MAP&VNF-DRA can get better performance and is benefit to improve the utilization ratios of network resources.3)Leveraging the model of deep belief network,this research designs a deep learning-based two-phase routing algorithm for SFC,which can not only get better network performance,but also enhance the time-efficiency of SFC routing computations.Based on the powerful characteristic abstraction and learning capacities,this research introduces deep learning technique into SFC routing problem and sets up a mathematical model for it aiming to minimize the end-to-end delay of the path for an SFCR,and further proposes the deep learning-based two-phase routing algorithm(DL-TPA).DL-TPA solves the problem with two process named as VNF selection and VNF chaining processes,and the VNF selection network and VNF chaining network are designed to take charge of these two processes,respectively.Moreover,in order to facilitate the time efficiency,DL-TPA optimizes the solution spaces of VNF selection network and VNF chaining network.Evaluation results present that,DL-TPA does better in time ef-ficiency than those of the rule-based compared algorithms,and can get better network performance as well.4)Considering the scenario of dynamic change of network load,this research uses deep reinforcement learning technique and designs a Double DQN-based VNF Placement Algorithm(DDQN-VNFPA),which can not only achieve efficient VNF placement,but also improve the network performance and utilization ratios of resources.This research consequently takes dynamic network load into account and sets up a mathematical model,where the objective is to optimize the VNF placement cost,VNFI running time and SFCR failure punishment given the change of SFC in a future time slot.Further,leveraging deep reinforcement learning,this dissertation proposes DDQN-VNFPA.Evaluation results show that,according to dynamic network load,DDQN-VNFPA can flexibly adjust and optimize the placement of VNFI in SDN/NFV-enabled networks and get better performance than those of the compared algorithms.