Research On Key Technologies Of Acceleration For Network Function Virtualization

In recent years,with the increase of network services in types and quantities,hardware devices in traditional networks have increased rapidly.Enterprises spend huge amounts of money on purchasing and deploying network devices.In addition,network devices usually come from different suppliers.It's necessary to equip with a large number of dedicated management personnel,resulting in the rapidly increasing of labor costs.In order to cope with the above problems,Network Function Virtualization(NFV)has emerged.Through virtualization technology,NFV deploys network functions on industry-standard high-capacity servers,switches and storage platforms,rather than dedicated hardware.It's very helpful for reducing the operating costs on network.The unified network equipment makes the network more versatile and adaptable.However,it also brings a series of problems.The capacity of the NFV system depends on the capacity of the universal platform.When dealing with some services that consume large computing resources,it is difficult to meet the service requirements through NFV platform.In order to overcome the above problems,this paper has put forward the following two aspects scheme.First,in order to improve the performance of the NFV architecture,this study proposes a flexible and efficient hardware acceleration scheme.For some of the functional modules originally deployed on general-purpose servers,it will be implemented on acceleration devices,the Field Programmable Gate Array(FPGA).Therefore,this study has built an FPGA acceleration platform.On this platform,a typical computing-intensive processing—hardware Data Encryption Standard(DES)encryption algorithm,and typical network-intensive processing——hardware Open Virtual Switch(OVS)were deployed.Through controlling the flow of business data to different acceleration modules.Business can be processed through different acceleration module.Test results have shown that the transmission delay of hardware OVS can reach 68 ns(under packet length of 64 Bytes).That is far smaller than software OVS and software-based acceleration scheme.Therefore,the hardware acceleration platform has great advantages in performance.In addition,new acceleration modules can be deployed through FPGA partial reconfiguration technology in this study.Flexible deployment of hardware acceleration modules and effectively savings hardware resources can be achieved.Second,in the actual network application scenario,the above hardware acceleration module needs to cooperate with other Virtual Network Functions(VNFs)which are deployed on general platform to form a functional service chain to provide services.The deployment location of the hardware acceleration module and the VNF may affect the end-to-end transmission of the service chain.So the overall delay of the service will be affected.Therefore,this study proposes a VNF migration algorithm for the NFV architecture with hardware acceleration modules.The virtual machine migration changes the actual deployment location of the VNF,thereby reducing the transmission delay of the service chain.In this study,the migration of virtual machines is planned by genetic algorithm.The simulation results show that the sum of transmission delays of all service chains is reduced by 86.4%.In summary,this study proposes a flexible hardware acceleration scheme based on FPGA,which effectively reduced the processing delay.A suitable virtual machine migration algorithm is proposed for the NFV architecture with hardware modules,which reduced the transmission delay.