Font Size: a A A

Packet Classification Research Based On Deep Reinforcement Learning

Posted on:2021-02-15Degree:MasterType:Thesis
Country:ChinaCandidate:Y R WangFull Text:PDF
GTID:2428330614450022Subject:Cyberspace security
Abstract/Summary:PDF Full Text Request
Packet Classification is a basic problem in the field of computer network.It aims to address the problem of how to efficiently analysis the incoming network packet and classify incoming network to separate flows and then choose actions to apply on the packet.as the evolution of the scale and performance of the network,the problem of packet classification has new requirements in efficiency and resource occupation.The traditional packet classification algorithm based on artificial rules and heuristic methods is not well suited on the problem at the current time.On the one hand,its classification performance is not satisfactory,on the other hand,it is restricted by poor generalization performance,which cannot meet the needs of the network environment of rapid iteration of network rules.In this paper,we propose a learning-based packet classification method which does not need human participation.We use the deep reinforcement learning technology to build the model,at the same time,we use the advantages of traditional heuristic packet classification method for reference,and realize end-to-end packet classification model construction.Through the experiment on the data set and comparison with the current mature packet classification method,we prove that the algorithm can greatly improve the classification performance and generalization performance of the packet classification problem.This paper describes the absorption and improvement of the traditional algorithm in detail,and explains the basic idea of applying traditional packet classification algorithm to the reinforcement learning field.At the same time,the design idea of the deep reinforcement learning method and the the design details of training algorithm are explained.Finally,we discuss the consistency between between reinforcement learning and packet classification task and why reinforcement learning algorithm can be successfully applied to packet classification task.In addition to the basic idea of the algorithm,this paper describes the implementation details of the algorithm in detail,especially the distributed training technology needed for the algorithm implementation on large-scale network,which provides a practical solution for the algorithm deployment in the real network environment.
Keywords/Search Tags:packet classification, reinforcement learning, decision tree, generalization performance, distributed training
PDF Full Text Request
Related items