On the cloud-based network traffic classification and applications identification service

Abstract

Recently, the need of traffic classification and applications identification has attracted numerous research efforts. Based on statistical attribute analysis, recent research studies employed machine learning algorithms for building traffic classifiers. The machine learning based traffic classification achieves high accuracy and becomes prominent scheme. This paper proposes the framework of cloud-based traffic classification service for sharing model and parallel classification. A training tool is designed for a PC to collect the mapping of “statistical information” and each application running in the PC. This statistical information is sent to the cloud for training. In the cloud, a database is designed to collect these information and a machine learning based training system is constructed. For traffic and applications classification service, the tool in the network of a company or a campus will also send the “traffic statistics” to the cloud, which then identifies these traffic by using virtual machines and returns the identified results. This service platform is scalable as the cloud platform is used and virtual machines can be rent and managed dynamically. Also based this training model, we have the opportunity to train/identify the network applications as complete as possible.