Hardware Offloading for Encrypted Traffic Classification Using P4-DPDK on SmartNICs

Author(s)

• Amith GSPN
• Akhil GBS
• Samia Choueiri
• Elie Kfoury, University of South CarolinaFollow
• Jorge Crichigno, University of South CarolinaFollow

Description

•The rise of encryption reduces network visibility, making encrypted traffic classification essential.

•Existing software-based solutions are flexible but struggle with high-speed traffic and heavy CPU usage.

•Hardware-based solutions offer high speed but are limited by fixed resources and simpler models.

•This work proposes a high-performance encrypted traffic classification (ETC) system.

•The system uses P4 and DPDK to achieve high-speed packet processing by bypassing the kernel stack.

•Packet processing and classification are offloaded to NVIDIA BlueField-3 SmartNIC Arm cores.

•Offloading avoids host CPU contention and improves overall efficiency.

•A Random Forest model classifies encrypted traffic using low-level packet header features.

•The system achieves up to 90% accuracy and completes inference in under 30 microseconds.

Publication Info

2026.

© 2026, Amith GSPN, Akhil GBS, Samia Choueiri, Elie Kfoury, & Jorge Crichigno