Research On Performance Optimization Of Network Function Virtualization Based On Container

With the evolution and development of network function virtualization(NFV)and edge computing,network functions(NFs)can be deployed on closer-to-end-user edge servers to reduce the amount of user-cloud data exchange and decrease end-to-end latency.Due to the limited resource capacity of edge servers,efforts are paid to develop lightweight container-based NFV platforms in edge network.However,in container-based NFV platform,multiple NFs usually share the same CPU core.These NFs will compete with each other for resource,which will degrade the performance of NFs,causing that the flow-level performance guarantee can not be provided.In this paper,we conduct empirical experiments on heterogeneous flows with different flow-level characteristics(i.e.,flow rate,packet size and packet loss rate),performance requirements(i.e.,latency and throughput)and different resources allocation.Then,by analyzing and summarizing the experimental results,we propose Finedge,a dynamic,finegrained and cost-efficient edge resource management platform for NFV network,which can provide NF service chain for each flow and assign the most cost-efficient CPU quota to each NF in order to provide flow-level performance guarantee.Finedge provides real-time flow monitor and NF manager to optimize the performance of container-based NFV platform.Finedge can automatically tune the most suitable CPU core with the most cost-efficient CPU quota by using designed resource allocation and scaling strategy for each NF.Through measurements,we collet a series of useful findings and insights from empirical experiments,which disclose the effect of NFs' resource allocation and their flowlevel characteristics on performance.Through extensive evaluation,we verify that proposed resource management schemes can efficiently handle heterogeneous flows with the lowest CPU quota and the highest SLA satisfaction rate as compared with other state-of-the-art resource management schemes.Furthermore,the proposed resource management platform supports real-time flow monitoring,elastic CPU quota allocation and scaling,and flexible NF migration among cores to handle the flow fluctuations.