论文地址:

Byte Segment Neural Network for Network Traffic Classification(2018-5-24)

EBSNN: Extended Byte Segment Neural Network for Network Traffic Classification(2021-8-2)

两篇论文,一篇是正常的字节编码,一篇是拓展字节编码。

EBSNN: Extended Byte Segment Neural Network for Network Traffic Classification

关键词:Recurrent neural network, traffific classifification, application identifification, website identifification

EBSNN结构

image-20211124182015657

处理步骤

Preprocessing

每个数据包是由Ethernet II header、the IPv4 header、 the TCP/UDP header and the payload构成的。

数据包的构成如table1。

image-20211124222750072

关于U和V的解释:

image-20211124222959923

所以,是如何进行preprocessing的呢?

  • Ethernet II header: only contains EtherType, source and destination MAC addresses, 直接放弃处理.
  • IPv4 header:放弃一些字段并将这些字段替换成0,
    • IP identifification (32 37 bits in the IPv4 header)
    • IP checksum(80 95 bits)
    • source IP address(96 127 bits)
    • destination IP address(128 159 bits).
  • TCP/UDP header:同IPv4,替换成0,
    • source port (0 15 bits)
    • destination port(16 31 bits)
  • Payload
    • 假设有M个字节,N个字节为一个字段,那么有[M/N]个字段。

举个栗子🍭

fig3中标红的都是替换成0的无用信息的字段;N=8,n=[Payload总字节数/N]

image-20211125002414543

Model