Research And Implementation Of Packet Classification Algorithm Based On GPU Acceleration

Packet classification algorithm has always been a hot topic in computer networks.As the basis of data forwarding by switching equipment in computer networks,its algorithm performance affects the data forwarding rate in the network.With the rapid development of the Internet,the data traffic and business demand in the network have soared,and the traditional network has been unable to meet the demand,so Software-defined networking(SDN)came into being,and the flow table responsible for packet forwarding in SDN contains at least 15 table entries and the flow table is huge,so the performance of the traditional 5-tuples packet classification algorithm cannot meet the requirements,so a new packet classification algorithm needs to be redesigned for the flow table.At present,due to more and more difficulties in Moore's law,parallel computing has been greatly developed,among which GPU accelerated heterogeneous parallel computing has outstanding performance,and has become a common parallel technology for emerging applications such as artificial intelligence and blockchain.Aiming at the problem of packet classification algorithm performance,this paper designs the corresponding packet classification algorithm by combining GPU heterogeneous parallel computing,and does the following work,and has achieved certain results.Firstly,this paper proposes a real-time dynamic packet classification algorithm based on dimension decomposition and the characteristics of each field in the network.This algorithm can effectively avoid scanning the whole rule set in the matching process and use classification search to improve the search efficiency of rule fields.Secondly,on the basis of the real-time dynamic package classification algorithm based on dimension decomposition mentioned above,a package classification architecture based on heterogeneous platform is designed.The parallel computing capability of GPU is utilized to design the pipeline structure of multithreaded search,so as to realize simultaneous batch matching rules,so as to improve the search speedof rule matching.Finally,we set up the experimental environment,to put forward the packet classification based on heterogeneous platform architecture and the architecture is based on decomposition of dimension of real-time dynamic packet classification algorithm is verified,and the experimental results show that under the same rule set size,in this paper,the real-time dynamic packet classification algorithm based on decomposition of dimension on the matching speed and scalability relatively this paper compared several algorithms of ascending to a certain extent,and the proposed packet classification based on heterogeneous platform architecture can effectively speed up the matching algorithm in this paper and improve the scalability of the algorithm.