Research And Implementation Of VPP-based NFV Performance Optimization Technology

Compared with traditional network function equipment based on proprietary hardware,network function virtualization technology not only improves hardware utilization,reduces initial investment and operating expenses,but also makes network functions easier to deploy and maintain,and can quickly adapt to constant changes.Therefore,it has received extensive attention.But compared with proprietary equipment,packet processing performance is still the main bottleneck of network function virtualization technology.Vector Packet Process(VPP)has become a feasible framework for network function virtualization with its fast packet process capabilities and flexible component expansion capabilities.However,the realization of network function virtualization based on the VPP framework is currently facing the challenges such as higher development costs of key network function components,lower throughput rates,and higher process latency.In response to these problems,this thesis studies the integration of Suricata-based deep packet inspection functional components in the VPP-based network function virtualization framework and optimizes the packet process performance of the framework.The specific work of this thesis is as follows:(1)In response to the lack of deep packet inspection in the VPP framework,the deep packet inspection engine Suricata is integrated into the VPP framework using a dynamic link library to implement a new deep packet inspection function component Suri VNF.Then the performance of Suri VNF was optimized from memory access,cache,multi-core and multi-thread concurrent programming,packet vector size,etc.Experimental results show that compared with the deep packet inspection function based on Suricata NFQ operating mode,Suri VNF has a 47.5 times higher throughput rate for 64-byte packets,and the average process latency drops to 5.34% of Suricata NFQ operating mode.After performance optimization,the throughput rate of Suri VNF process 64-byte packets increased by 16.9%,and the average process latency was reduced by 17.98%.(2)Aiming at the problem of low IP lookup performance of VPP under the condition of large-scale routing table,an IP lookup performance optimization method based on cuckoo filter is designed.First,a new cuckoo filter OCF filter is designed,and the priority strategy and modulo operation are used to optimize the update operation and space consumption of the cuckoo filter,and then it is applied to the IPv6 longest prefix matching of VPP to achieve the new IP lookup component OCFVNF combining VPP's packet vectorization process and cache locality to optimize the lookup rate of the OCF filter.Experimental results show that OCFVNF can increase the IP search rate of VPP by 2.06 times.In the worst case,the space consumption of the OCF filter is reduced by 50% compared with the cuckoo filter,and the update time is reduced by 31.86%,which effectively improves the IP search performance of VPP.Optimized the key performance indicators of the VPP-based network function virtualization framework under the condition of large-scale routing tables.(3)Based on the previous work,the extended VPP network function virtualization framework e VPP is designed.The e VPP framework extends the deep packet inspection function component Suri VNF and the IP lookup component OCFVNF.The experimental results show that under the condition of large-scale routing table,the throughput rate of64-byte packets of the e VPP framework is 69.73 times higher than that of Suricata NFQ mode,and 19.25% higher than that of Suri VNF.In short,the VPP-based network function virtualization framework implemented in this thesis uses the integration of Suricata to expand the deep packet inspection function and optimizes the throughput,packet process latency and IP lookup performance in the program design mode,software and hardware configuration parameters,and filter designed,which improves the practicability of the network function virtualization framework.