Résumé
This paper introduces Nxcap, a unified network traffic floworiented capture format designed to bridge the gap between descriptive statistics and packet capture files (Pcap) for network intrusion detection systems (NIDS) benchmarking. Nxcap represents network traffic as a collection of flows with associated packets, using a minimal set of features that enable both Pcap reconstruction and the extraction of flow-level descriptive statistics. This format addresses the limitations of existing datasets, which often suffer from inconsistencies, lack of standardization, and errors due to misconfigurations or biases introduced during data collection. We present libnxcap, an open-source library that captures network traffic in the Nxcap format, facilitates flow labeling, and includes mechanisms to verify and correct data at the time of capture. A key contribution is the advanced deduplication engine, which operates at the flow level and accounts for protocol behavior, revealing significant duplication issues in popular datasets like UNSW-NB15. Experiments demonstrate that Nxcap reduces dataset storage requirements by over 98% compared to raw Pcap files, while ensuring data quality and compatibility with existing tools. This work aims to provide the research community with a standardized, efficient, and reliable format for network traffic representation, enhancing the development and evaluation of NIDS.