Tesi etd-09112025-110228
Link copiato negli appunti
Tipo di tesi
Corso di Dottorato (D.M.226/2021)
Autore
ABU BAKAR, RANA
URN
etd-09112025-110228
Titolo
Data Processing Units for Next-Generation Networks: Evaluation across User Plane Function, High-Performance Computing, and Cybersecurity Workload
Settore scientifico disciplinare
ING-INF/03
Corso di studi
Istituto di Tecnologie della Comunicazione, dell'Informazione e della Percezione - Ph.D. in Emerging Digital Technologies
Commissione
relatore Prof. CASTOLDI, PIERO
Presidente Prof. GREGORIO PROCISSI
Membro Dott. KYRIAKOS VLACHOS
Membro Prof. SAMBO, NICOLA
Tutor Dott. CUGINI, FILIPPO
Presidente Prof. GREGORIO PROCISSI
Membro Dott. KYRIAKOS VLACHOS
Membro Prof. SAMBO, NICOLA
Tutor Dott. CUGINI, FILIPPO
Parole chiave
- SmartNIC
- DPU
- Hardware
- Offload
- Machine Learning
- AI
- Wirespeed AI
- Networks
- GNN
- DPI
- DOCA Flow
- Flow Tracking
- DDoS Detection
Data inizio appello
03/11/2025;
Disponibilità
completa
Riassunto analitico
The rapid evolution of next-generation networks demands high performance, scalability, and security. However, conventional CPU-based systems face fundamental bottlenecks in terms of throughput, latency, and energy efficiency, which limit their ability to support 5G networking, HPC, and cybersecurity workloads. This thesis aims to evaluate the role of programmable Data Processing Units (DPUs) as a unifying hardware platform for accelerating workloads across 5G data networks, high-performance computing (HPC), and network security.
The research investigates three complementary directions. For 5G, user-plane functions such as GTP encapsulation, QoS enforcement, and DDoS mitigation are offloaded to NVIDIA BlueField DPUs using the DOCA framework. For HPC, a comparative study of BlueField-2 and BlueField-3 is conducted using RDMA micro-benchmarks to assess bandwidth, latency, and message-rate performance. For security, the thesis introduces IDS-NGNN, a nested graph neural network for intrusion detection, and proposes PUF-based authentication schemes for autonomous swarm systems.
The findings show that DPUs can sustain line-rate packet processing up to 400 Gbps in 5G scenarios while reducing latency and improving scalability compared to software-only UPFs. In HPC contexts, comparative benchmarking highlights clear trade-offs between BlueField-2 and BlueField-3 in terms of throughput and latency, underscoring the suitability of DPUs as data movement accelerators. In security, the combination of IDS-NGNN and PUF-based mechanisms demonstrates that AI and hardware primitives can provide robust, low-latency protection against evolving cyber threats.
This thesis concludes that DPUs are a cornerstone technology for the 6G era and beyond, enabling infrastructures that are not only faster and more efficient but also inherently more secure and resilient. By bridging advances in 5G networking, HPC, and cybersecurity, the work establishes DPUs as critical enablers of future high-performance, secure digital ecosystems.
The research investigates three complementary directions. For 5G, user-plane functions such as GTP encapsulation, QoS enforcement, and DDoS mitigation are offloaded to NVIDIA BlueField DPUs using the DOCA framework. For HPC, a comparative study of BlueField-2 and BlueField-3 is conducted using RDMA micro-benchmarks to assess bandwidth, latency, and message-rate performance. For security, the thesis introduces IDS-NGNN, a nested graph neural network for intrusion detection, and proposes PUF-based authentication schemes for autonomous swarm systems.
The findings show that DPUs can sustain line-rate packet processing up to 400 Gbps in 5G scenarios while reducing latency and improving scalability compared to software-only UPFs. In HPC contexts, comparative benchmarking highlights clear trade-offs between BlueField-2 and BlueField-3 in terms of throughput and latency, underscoring the suitability of DPUs as data movement accelerators. In security, the combination of IDS-NGNN and PUF-based mechanisms demonstrates that AI and hardware primitives can provide robust, low-latency protection against evolving cyber threats.
This thesis concludes that DPUs are a cornerstone technology for the 6G era and beyond, enabling infrastructures that are not only faster and more efficient but also inherently more secure and resilient. By bridging advances in 5G networking, HPC, and cybersecurity, the work establishes DPUs as critical enablers of future high-performance, secure digital ecosystems.
File
| Nome file | Dimensione |
|---|---|
| Thesis_F...rsion.pdf | 3.57 Mb |
Contatta l'autore |
|