- 期刊
A Spark-based Method for Identifying Large-scale Network Burst Traffic
摘要
The identification of network traffic plays a vital role in ensuring the safe, stable, and efficient operation of the network. To identify large-scale network burst traffic efficiently, a distributed convolutional neural network method based on Spark is proposed. The 'Raw' data in the TCP protocol is extracted as inputs, and CLR-Distributed-CNN (a distributed convolutional neural network with a cycle learning rate) is used to distinguish network traffic. The accuracy rate of this method reaches 90.417%. Finally, Distributed-RF (a distributed random forest) and EXP-Distributed-CNN (a distributed convolutional neural network with exponential decay of learning rate) are designed to compare with the new method. The accuracy of EXP-Distributed- CNN is 88.167% and that of Distributed-RF is 81.433%. Therefore, the experimental results demonstrate its feasibility and validity.