Research On Optimizing Service Orchestration Based On Network Function Virtualization

Facing the problems of rigid network structure,low resource utiliza-tion and complex management based on the traditional dedicated hardware network devices,Network Function Virtualization(NFV)technology has been developed,which transforms the dedicated hardware-based network function devices into Virtual Network Functions(VNFs)devices running on general-purpose servers in the form of virtual machines,containers or processes through virtualization technology.Though the deployment of VNFs shows many advantages,such as fast and flexible service deploy-ment,efficient resource utilization,and low overhead operation manage-ment,it brings multiple new issues and challenges.VNFs usually work as a cluster of which the total processing capacity can be dynamically scaled for high resource utilization when facing fluc-tuating service traffic.However,it remains an urgent issue to be addressed to design an efficient traffic scheduling mechanism for the dynamic scaling of VNF clusters.In addition,though NFV makes the deployment of SFCs flexible,how to design the algorithms for SFCs deployment remains open.Finally,backup nodes have to be deployed for the SFCs,thereby improv-ing the overall availability.Therefore,it becomes increasingly important to allocate backup nodes efficiently to meet the availability requirements of the services.In view of the above issues and challenges,this paper fo-cuses on three specific aspects,namely,flow scheduling and VNFs scaling,SFCs deployment,and backup allocation for SFCs.The designed mecha-nisms and algorithms will drive the service orchestration toward automatic,continuous,real-time,and optimized directions.The specific work of this paper includes the following four points.1.Joint optimization of flow scheduling and VNF scaling in VNF clusters.The dynamic change of service traffic makes it necessary to realize dynamic scaling within the VNF cluster,and schedule the service traffic reasonably at the same time.However,the hash-based scheme and the flow-table-based scheme cannot balance the flows ef-ficiently.For this reason,as a framework for joint flow scheduling and VNF scaling,Hieff is designed to differentiate large and small flows according to the flow rate,that is,Hieff uses flow tables to schedule large flows,and consistent hashing to schedule the small flows.In addition,Hieff jointly optimizes the flow scheduling and VNF scal-ing strategies to maintain the load of all active instances in the cluster nearby a predefined target value,thereby reducing the flow migration and instance management overhead.2.Joint optimization of SFC composition,placement and assign-ment.Before being deployed,SFC first needs to be composed based on the service requirements.However,no work has been done to jointly optimize the composition,placement and assignment of SFCs.To fill this gap,the issue of SFC-CPA is proposed in this paper,which introduces SFC dependency matrixes and generation matrixes,and models the SFC composition,placement and assignment co-optimization problem as a 0-1 linear programming.The proposed Jcap algorithm is available for an efficient improvement of the re-source utilization of VNF instances,and the reduction of the physical link bandwidth consumption.Besides,the Jcap algorithm is efficient with its performance close to the optimal solution.3.Optimizing SFC deployment with stringent end-to-end delay guarantees.Latency-sensitive services,such as autonomous driving,telemedicine,etc.have more stringent requirements on the end-to-end delay of the SFC.However,the current SFC deployment efforts can-not guarantee that every packet can be transmitted within the calcu-lated delay.To solve this problem,the issue of DetSFC deployment,which aims to maximize the volume of the accepted services while guaranteeing that the end-to-end delay of the accepted service strictly meets the requirements;in another word,the delay of each packet passing through the SFC can be guaranteed.For this purpose,DetSFC derives an end-to-end delay bounds for packets passing through the SFC based on network calculus theory,and verifies the accuracy of the derived results through experiments.According to the results of network calculus,the proposed algorithm JRRA achieves the objec-tives of DetSFC by jointly optimizing SFC routing and VNF instance resource allocation.4.Resource-aware backup allocation for a chain of VNFs.The avail-ability of VNFs declines compared with the hardware-based network function devices.Therefore,the overall availability of the SFC is re-duced,and the backup nodes need to be deployed.Different VNFs have different availability parameters,and require different amount of resources.How to allocate backup nodes for the SFC has a great impact on the overall availability of the SFC and the resource con-sumption of backup nodes.However,the existing algorithms ignore the heterogeneous resource requirements of VNFs when designing backup allocation scheme.In this case,a Resource-Aware Backup Allocation(RABA)scheme for SFCs is proposed,which designs the methods for SFCs overall availability calculation for both dedicated and shared backup scenarios,optimizes the backup node allocation,and finally achieves the goal of minimizing resource consumption of backup nodes while satisfying the overall availability requirements of SFCs.