CESNET-TLS-Year22

Overview

The CESNET-TLS-Year22 dataset is an extensive year-long collection of TLS-encrypted traffic containing 507 million flows across 180 web services. The dataset is annotated using the same process as CESNET-TLS22 and CESNET-QUIC22, providing a comprehensive resource for studying traffic classification over extended time periods.

Dataset Metadata

Dataset Characteristics

Sampling Strategy

Due to the substantial volume of TLS traffic collected over a full year, uniform sampling was applied:

• Sampling Ratio: 1:10
• Purpose: Make dataset size manageable while preserving temporal patterns
• Method: Uniform random sampling across all time periods
• Preservation: Maintains representative distribution of traffic patterns

Included Data

Each flow record includes:

• Flow Features: Standard IP flow attributes (bytes, packets, duration, etc.)
• SNI Domain: Extracted Server Name Indication from TLS handshake
• Timestamp: Precise capture time for temporal analysis
• Service Label: Annotated web service class (180 classes)

Note: Unlike CESNET-QUIC22, this dataset does not contain background traffic classes—all flows are labeled web services.

Research Applications

The CESNET-TLS-Year22 dataset’s year-long span makes it particularly suitable for:

1. Traffic Classification Robustness

• Temporal Stability: Evaluate classifier performance over extended periods
• Seasonal Patterns: Study how traffic changes across months and seasons
• Evolving Behaviors: Assess impact of service updates and protocol changes

2. Model Retraining Strategies

• Retraining Frequency: Determine optimal model update intervals
• Incremental Learning: Test continuous learning approaches
• Transfer Learning: Evaluate knowledge transfer across time periods

3. Incremental Class Learning

• New Class Addition: Add emerging services without full retraining
• Class Evolution: Handle changes in existing service behaviors
• Memory Efficiency: Minimize computational requirements for updates

4. Few-shot Learning

• Underrepresented Classes: Improve performance on rare services
• Emerging Applications: Classify new services with limited samples
• Cold-start Problem: Handle newly deployed web services

Comparison with Related Datasets

How to Cite

@article{luxemburk2024long,
  title={Long-term Traffic Classification using Self-supervised Representation Learning},
  author={Luxemburk, Jan and Hynek, Karel and {\v{C}}ejka, Tom{\'a}{\v{s}}},
  journal={Computer Networks},
  year={2024},
  publisher={Elsevier}
}

Download

[1] Luxemburk, J., Hynek, K., & Čejka, T. (2024). CESNET-TLS-Year22: One-Year TLS Network Traffic Dataset [Data set]. Zenodo.
Download link to be added