Uma abordagem inovadora para classificação de tráfego em tempo real em ambiente de redes definidas por software (SDN)

• Autor(a) principal: Klenilmar Lopes Dias
• Data de Publicação: 2019
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da UFMG
• Texto Completo: http://hdl.handle.net/1843/37466; https://orcid.org/0000-0003-2769-3647
• topic: Classificação de tráfego; Redes definidas por software; Aprendizado de máquina; Naive bayes; Streaming de vídeo; Engenharia elétrica; Aprendizado do computador; Teoria bayesiana de decisão estatística; Redes de computadores

The growing demand for high-speed transmission rates on recent years has attracted research in new mechanisms for network traffic characterization and classification. Their inadequate treatment degrades the performance of important operational schemes, such as Network Survivability, Traffic Engineering, Quality of Service (QoS), and Dynamic Access Control, among others. The most common methods for traffic classification are Deep Packet Inspection (DPI) and port based classification. However, those methods are becoming obsolete, as increasingly more traffic is being encrypted and applications are using dynamic ports or ports originally assigned to other popular applications. In this same direction, the problem arises not only in sustaining the future volume of traffic in Data Centers, but also in sustaining the current traffic and its oscillations with this new demand for high-speed transmission rates required by Internet services. One of the main problems generated by the increase in volume is the efficient use of the throughput, to attend the requisitions. In addition, in recent years the Software Defined Networks (SDN) paradigm has brought new opportunities for centralized network management. Thus, this work presents a SDN traffic classifier, with an innovative approach to real-time network traffic classification, for use in Traffic Engineering solutions in Data Centers. The classifier adopts a new approach for the relaxation of the hypothesis of independence between the attributes of the Naive Bayes algorithm. The results show that the proposed module is a promising alternative for use in real-time scenarios.