POSMAC: Powering Up In-Network AR/CG Traffic Classification with Online Learning

TL;DR

POSMAC addresses the challenge of AR/CG traffic classification under encryption and dynamic port usage by combining PCAP-driven data augmentation with an online learning loop executed on a DOCA-enabled BlueField-3 DPU. The architecture comprises four interlinked components—PCAP Pool, Traffic Classifier, Application Servers, and Online Trainer—that perform feature extraction, classification at the edge, flow routing, and continual retraining to maintain accuracy. The approach demonstrates a complete in-network, online-learning pipeline, achieving 95.7% accuracy in simulation with transfer learning pushing results toward 100% and providing real-time performance metrics. This work enables more accurate, scalable AR/CG traffic classification at near line-rate in in-network dataplanes, with a clear path toward hardware validation on BlueField-3. Practical impact includes reduced deployment costs and improved adaptability to evolving traffic patterns through continuous learning.

Abstract

In this demonstration, we showcase POSMAC1, a platform designed to deploy Decision Tree (DT) and Random Forest (RF) models on the NVIDIA DOCA DPU, equipped with an ARM processor, for real-time network traffic classification. Developed specifically for Augmented Reality (AR) and Cloud Gaming (CG) traffic classification, POSMAC streamlines model evaluation, and generalization while optimizing throughput to closely match line rates.

Figures (1)

• Figure 1: POSMAC Architecture: Dotted lines illustrate online learning data flows — black for feature extraction, blue for class labeling, and red for transferring the retrained ML model.