Payload-based packet classification and its applications in packet forwarding pipeline

Abstract

As the number of protocols and applications is increasing, the need for accurate and dynamic identification methods becomes crucial for many applications such as security and QoS. The SDN and programable data plane architectures changed the way of designing the network and how to implement networking functions. In our work we will use that advance in the architecture to solve the challenge of packet classification. Although there are other payload-based techniques that tackle this problem, but each method has some limitations and it is used in specific scenarios and usually compromise some advantages over some others. The main approaches are either using regular expressions, statistical properties, or some app signature extracted by complex natural language processing techniques. Our approach is aimed to be fast, simple, easy to be implemented, accurate and learn new protocols. To achieve this, we introduced a new approach of keyword generation based on (position, byte) model which insures the uniqueness, in which a protocol is not defined by the statistical properties of the first few bytes of the packets and not all the first few bytes are used. Our work currently is in its mid-stage and it is promising technique that will reduce the complexity and processing time and achieve higher order of accuracy.