GPU-based SDN network packet classification method and system

Abstract

The invention discloses a GPU-based SDN network packet classification method and system, and the method comprises the steps of: enabling an SDN switch to create a two-dimensional array, and transmit initial data of a rule set, a matching kernel, a calculation kernel and the two-dimensional array to a GPU; enabling the GPU to receive and store the initial data of the SDN switch; enabling the SDN switch to analyze a packet header field in the SDN network packet and send the packet header field to the GPU; carrying out parallel computing on a plurality of threads in the GPU, and carrying out kernel matching and kernel computing; matching the packet header field with the rule set field, and storing a matching result into a two-dimensional array; performing logical operation on each row of datain the two-dimensional array, and screening an operation result to obtain a rule set number; screening the rule set number with the highest priority and sending the rule set number to the SDN switch;and processing the SDN network packet according to the data processing rule corresponding to the rule set number with the highest priority. According to the method, the characteristic of parallel matching of the data packets is utilized, the parallel computing capacity of the GPU is combined, and it can be guaranteed that the updating rate is not affected while rapid matching is achieved.

Description

technical field [0001] The invention relates to the technical field of SDN network packet classification, in particular to a GPU-based SDN network packet classification method and system. Background technique [0002] Software Defined Networking (SDN) adopts the idea of ​​centralized control and distributed forwarding. At present, the mainstream forwarding protocol adopted by switches in SDN network is OpenFlow protocol. The size of the flow table in the network has increased sharply, and the traditional packet classification system is far from being suitable for SDN networks. Traditional networks only use their best efforts to transmit data packets to their destinations, so data packets may be delayed or lost. [0003] The packet classification system is a way to extract multiple fields in the header of the data packet. Taking the traditional five-tuple as an example, the source / destination IP address, source / destination port number and protocol type fields are matched with...

AI Technical Summary

Problems solved by technology

[0003]The packet classification system is a kind of extraction of multiple fields in the data packet header, taking the traditional five-tuple as an example, source / destination IP address, source / destination port number and the protocol type field, match the rules in the rule set according to their respective matching methods, and retrieve and find the highest priority rule, so as to execute the rule action system. For the current packet classification system commonly used in the SDN network, it includes: linear search system , tuple space search system and tree structure-based search system, among them, the linear search system can only match one rule at a time, although the system flow is simple, but when the rule set is large, the matching speed is too slow; although the tuple space search system Compared with the linear search algorithm, the search speed is improved, but the preprocessing of the rule set in the grouping process is relatively slow, and the process of calculating the mask is also time-consuming, requiring a high-performance CPU to satisfy the calculation; although the search algorithm based on the tree structure is extremely large The search speed is improved, but the algorithm for constructing the search tree structure is complex and the process is very slow, which is not conducive to the update of the rule set

[0004] To sum up, the commonly used packet classification system in SDN network currently has the problem of slow matching speed, or the matching speed becomes faster but the update speed of the rule set is too slow

Images