Defesa – Dissertação de Nicolas Ribas

Aluno(a): Nicolas Kolling Ribas
Orientador(a): Lisandro Zambenedetti Granville
Coorientador(a): Marco Aurélio Spohn

Título: High-Speed In-Kernel Network Periodicity Detection: A Discrete Wavelet Transform Implementation via eXpress Data Path
Linha de Pesquisa: Arquiteturas, Protocolos e Gerência de Redes e Serviço

Data: 02/09/2026
Hora: 14:00
Local: Esta banca ocorrerá de forma remota. Acesso público disponibilizado pelo link https://conferenciaweb.rnp.br/sala/lisandro-zambenedetti-granville.

Banca Examinadora:
-Carlos Raniery Paula dos Santos (UFSM)
-Luciano Paschoal Gaspary (UFRGS)
-Ricardo Jose Pfitscher (UFSC)

Presidente da Banca: Lisandro Zambenedetti Granville

Resumo: Periodicity in network traffic often indicates anomalies; yet, analyzing high-speed streams is challenging due to the volume of data. While signal processing techniques such as the discrete wavelet transform (DWT) are effective, their real-time application in the network data plane remains non-trivial. This dissertation presents the design and implementation of an online DWT-based method for detecting periodicity in network traffic using the eXpress Data Path (XDP). To this end, we develop an algorithm that bypasses key constraints of the extended Berkeley Packet Filter (eBPF), such as the lack of floating-point arithmetic, and performs DWT decomposition entirely in the XDP data plane. The algorithm operates on a time-series signal generated from packet-per-second counts. We validate our approach by applying it to real-world, security-related traffic captures containing known periodic patterns, which our approach successfully identifies. We also assess the performance of our solution by comparing its throughput against a baseline scenario on high-performance hardware, specifically a 100 Gbit/s BlueField-2 DPU, thereby demonstrating its suitability for high-traffic environments.

Palavras-Chave: Data Plane Discrete Wavelet Transform eBPF Network Security Periodicity Detection Traffic Analysis