TRAFFIC CLASSIFICATION IN MODERN NETWORKS CHALLENGES AND LIMITS

Abstract

Traffic classification is essential for QoS, security, and billing, but modern networks are dominated by encrypted traffic, dynamic ports, and rapidly evolving protocols, which reduce the effectiveness of DPI and port-based methods. This paper outlines the classification problem, key evaluation metrics, and two resource-efficient directions for high-speed networks: sampling fewer packets/flows and selecting only the most informative features. Remaining challenges include real-time latency constraints, scarce labeled datasets, difficult labeling, unknown applications, and ambiguity caused by shared cloud infrastructures.