Fine-grained monitoring and classification of interactive media network traffic

• Autores: Tianhua Chen
• Directores de la Tesis: Elans Grabs (dir. tes.), Aleksandrs Ipatovs (dir. tes.), María Dolores Cano Baños (codir. tes.)
• Lectura: En la Riga Technical University ( Letonia ) en 2026
• Idioma: inglés
• Número de páginas: 143
• Enlaces
  • Tesis en acceso abierto en: ebooks.rtu.lv
• Resumen

  • Network Traffic Monitoring and Analysis (NTMA) plays a critical role in assessing network performance and optimizing resource utilization. In today's heterogeneous network environments, interactive media applications consume a substantial portion of network traffic. Emerging services such as cloud gaming, virtual reality (VR)/metaverse, and live streaming with 4K/8K resolution and 60/120 frame rates content are rapidly expanding and significantly increasing traffic volume. In conventional Network Traffic Classification (NTC) tasks, traffic generated by these applications is often broadly categorized as video traffic. These coarse labeling limits precision and leads to underrepresentation in many public Internet traffic datasets. Moreover, even traffic from the same application or service may display different characteristics under varying network conditions in heterogeneous environments. To address this gap, the doctoral thesis aims to achieve fine-grained monitoring and classification of network traffic generated by interactive media applications across diverse network conditions. The classification is conducted at three granularity levels: time series, flow, and payload. This approach enhances understanding of interactive media traffic patterns and their roles in complex network environments. It also contributes to traffic behavior modeling for NTC tasks and extends to broader NTMA applications. To support this goal, four categories of interactive media applications were selected. Corresponding datasets were created at each granularity level. Five deep learning and ensemble models were developed and evaluated against three state-of-the-art classifiers to assess classification performance.