Research On Resource Mapping Model Of Network Function Virtualization On Clearwater Platform

Network Function Virtualization(NFV)shows agility and scalability in industry when de-ployed in cloud computing infrastructure,but the performance of softwarized network service still impedes its popularization.Most of the network service in cloud run as service function chain(SFC)in Standard High Volume Server,consisting of virtual machines(VMs)in tandem.Each VM runs a dedicated virtual network function(VNF)in SFC.The hyper-visor treats these VM with running VNF as single instance to maintain.When a certain SFC is called for serving,the resource manager lacks the recognition of details on the connections between running VFs and separately allocate the resource which consequently lead to low efficient resource utiliza-tion for SHVS.This kind of resource allocation will result in performance fluctuation in SFC.How to optimize the SHVS resource efficiently for chain-style NFV application poses a great challenge for us.IMS(IP Multimedia Subsystem)is an architectural framework for delivering IP multime-dia services which has been broadly applied in communication solutions.Inspired by the idea of NFV,the service providers and open source communities have released a bunch of projects combined with the support of virtualization techniques and cloud computing.The IMS system comprises several function groups which are linked by specific orders to fulfill the service re-quirements.This form of organization serves as the service function chain.To deploy in the virtualized environment,the IMS system also meets the challenges of performance fluctuation and resource utilization.To resolve this problem,in this paper,we propose a new resource constraint abstraction considering the chain-style NFV workload with underlying characteristics and design a greedy algorithm based on our model to map resource.We implement our framework in real IP Mul-timedia IMS platform called Clearwater.First,we analyze core services of Clearwater and conclude the service function chains of these services.Second,we use the Hardware Prober to detect the information of running environment,and take these information as the input of the model and algorithm.The Dynamic Coordinator leverages the information to generate mapping strategies and deploys the optimized mapping strategies in real platform.At last,the framework is tested and compared to existing generic methods.By the exami-nation statistics,our framework can decrease 40%average IMS service latency and improve the throughput within the simulative SFCs up to 200%.