Independent comparison of popular DPI tools for traffic classification

摘要: Deep Packet Inspection (DPI) is the state-of-the-art technology for traffic classification. According to conventional wisdom, DPI most accurate classification technique. Consequently, popular products, either commercial or open-source, rely on some sort of However, actual performance still unclear research community, since lack public datasets prevent comparison and reproducibility their results. This paper presents a comprehensive 6 well-known tools, which are commonly used in literature. Our study includes 2 products (PACE NBAR) 4 open-source tools (OpenDPI, L7-filter, nDPI, Libprotoident). We studied various scenarios (including packet flow truncation) at different levels (application protocol, application web service). carefully built labeled dataset with more than 750K flows, contains from applications. Volunteer-Based System (VBS), developed Aalborg University, guarantee correct labeling dataset. released this dataset, including full payloads, community. believe could become common benchmark validation network classifiers. results present PACE, tool, as solution. Surprisingly, we find that such nDPI Libprotoident, also achieve very high accuracy.