Traffic Refinery
Cost-aware data representation analysis system
Overview
Traffic Refinery is a proof-of-concept system implemented in Go that monitors network traffic at 10 Gbps and transforms it in real-time to produce various feature representations for machine learning.
Features
- Real-time traffic processing at 10 Gbps
- Flexible, extensible network data representations
- Systems cost profiling for different representations
- Flow categorization using DNS domains or IP prefixes
- Customizable feature extraction with configurable frequency
Resources
- Website: traffic-refinery.github.io
- GitHub: github.com/traffic-refinery
- Paper: ACM POMACS
Use Cases
- Video streaming quality inference
- Malware detection
- Network performance analysis
- Any ML task requiring cost-aware traffic representation