Research On The Hardware Acceleration Mechanism For SDN/NFV

In recent years,Software-defined Networking(SDN)and Network Function Virtualization(NFV)technologies have broken the closed rigid structure of existing networks,enhancing the scalability of network functions and the flexibility of network configurations,which have received extensive attention from academia and industry.However,in the SDN/NFV architecture,the performance of various virtualized network functions(VNFs)implemented in software such as firewalls and load balancers is seriously degraded,which cannot meet the network business requirements with high-rate and delay-sensitive scenarios.To this end,researchers began to use hardware acceleration mechanism to offload some data processing tasks to various types of acceleration hardware such as FPGAs,network processors and so on,improving the data processing performance of VNF through co-processing between software and hardware.The hardware acceleration mechanism breaks through the performance bottleneck of the traditional SDN/NFV architecture,and has important academic research significance and practical application value.This paper relies on the National 863 Program topic "Software-Defined Network Architecture and Key Technology Research",focusing on the hardware acceleration mechanism for SDN/NFV.Aiming at the problem of tight coupling design between existing hardware acceleration structure and VNF and the lack of unified deployment and orchestration of hardware acceleration resources,firstly a general hardware accelerator structure that can dynamically and flexibly load VNF is designed.Then,based on such structure,an optimized deployment strategy for all network acceleration hardware and a service chain mapping method supporting hardware acceleration are proposed respectively.These efforts improve the data processing performance of SDN/NFV networks.The main innovations are as follows:1.A general hardware accelerator(GHA)structure,which realizes dynamic and flexible loading of VNFs,is designed to solve the problem of tight coupling design between the existing hardware acceleration structure and VNF.GHA structure can support a variety of VNFs with a programmable parser and an action processing unit.Secondly,using run-time reconfiguration,GHA deploys different accelerators of VNF dynamically and independently.Thirdly,the discrete particle swarm optimization(DPSO)algorithm is used to optimize the allocation of acceleration resources in data centers.Finally,the GHA prototype experiment results show that the VNF processing speed of the structure is increased by 2 times compared with the existing DPDK acceleration scheme.2.A two-stage acceleration resource deployment(TARD)strategy for improving the utilization of hardware acceleration resource is designed to solve the the problem of optimizing the deployment of limited hardware acceleration resources in the network.In the first phase,considering the physical attributes and topological positions of the switch devices,aiming at adapting to the traffic carried by the forwarding node,the topological potential is used to quantify the acceleration resources deployment in the switches.In the second phase,using the cooperation of the acceleration resources,the deployment of the server accelerators is completed.Finally,simulation results show that compared with the single-attribute acceleration resource deployment algorithm(SARD),the proposed strategy can optimize the deployment for the acceleration resources and improve the total traffic handled by acceleration resource and the utilization of acceleration resource by 14.2% and 10.9% respectively.3.Aiming at realizing an optimal bearer of the service function chain,a service function chain placement method with multiplexing acceleration resource priority is proposed to solve the problem of efficient orchestration of hardware acceleration resources in the whole network.The algorithm prioritizes the reuse of acceleration resources in the switch under the optimal placement of VNF without acceleration to commercial servers.And the mapping correlation between hardware acceleration resources and VNF is flexibly adjusted according to the requirements of network services.Prototype experiments and simulation results based on acceleration table and acceleration card joint acceleration show that compared with the existing service function chain placement methods,the algorithm improves the service traffic contained in the network and placement success rate by 13.6% and 9.9% in the case of meeting the high processing throughput needs.