Research Of Packet Classification Technology Based On Multi-core NPU

Packet classification technique plays a more important role in varies of network equipments and applications.For instant,fire wall,network information detection,next generaion routers,security gateway and so on.The performance has tremendous impact on the quality of service.With the network speed becomes more and more quick,the need of efficient packet classification algorithms also becomes more urgent. Currently,researchers have already developed several hardware and software algorithms.And these algorithms can satisfy some application scene.Since the complexity of IP packet classification problem in theory, Existing hardware and software algorithms have their own shortcomings exist,there is not an ideal algorithm that satisfy requiement in all cases till now.and there are not many researches based on multi-core NPU.The needs for efficient packet classification algorithms are very urgent in pratical application.This paper presents a multi-core-NPU based MCN-EGT algorithm,which is also based on exiting single-thread EGT algorithm,it uses multiple thread to accelerate the speed of classification. Multiple thread can hide memory access latency,and can parallel some steps in packet classification, so as to achieve the effect of accelerating.The main improvements include initial classification to the rules and packets ,and hide the update time of the rules.And this paper describe the algorithm in detail in pre-processing,matching and updating stages.This paper also proposed a packet classification algorithm based on function distribution.It takes 5-tuple(source and destination IP, source and destination port, protocol)for intance,devide its 104 bits into 13 parts by byte,and take multi-threaded execution in the pre-processing, matching and updating stages,so as to maximize the use of multi-core parallelism.And threr is little association between threads,therefore,the communication and synchronization cost is also small.Finally,we tested the MCN-EGT algorithm and packet classification algorithm based on function distribution.The test results show that the MCN-EGT algorithm is 2-3 times better than the EGT algorithm.The match performance of packet classification algorithm based on function distribution is better than that of MCN-EGT algorithm.In the update speed aspect,MCN-EGT algorithm performed best,and the packet classification algorithm based on function distribution followed. The scalabilit of packet classification algorithm based on function distribution is best of the three,and it can change the length with the field.This paper studies the problem of packet classification status and shortcomings,and proposes two algorithm based on multi-core NPU.Their performance are improved over the previous algorithm.But there is still much to be optimized,improved and researched.Further work will focus on further improving the performance.