Description
Traffic Refinery is an open-source research system for cost-aware network traffic analysis and machine learning. It was built to measure how the choice of network data representation affects both the systems cost and the performance of models trained on it, since representations are often chosen in advance without regard to how models will use them. The system is written in Go and has three parts: a traffic categorisation module that labels flows as services using DNS domains or IP prefixes you provide, a packet capture and processing module that tracks flow state in a cache, and an aggregation and storage module that queries that cache for features and statistics about each flow. The project supports cost profiling, offline analysis of pcap files, and built-in benchmarking tools, and its documentation covers installation, usage, configuration and extending features. The code is on GitHub under a licence file, and the project lists a publication with a citation. It suits network researchers and engineers building machine learning pipelines on traffic data.…
Highlights
- Cost-aware network traffic analysis
- Labels flows by DNS domain or IP prefix
- Offline analysis of pcap files
- Cost profiling and benchmarking
- Written in Go
Ratings & Reviews
No ratings yet
Be the first to rate Traffic Refinery.
People also visit
Traffic Refinery FAQ
What is Traffic Refinery?
An open-source system for analysing network traffic for machine learning, and for measuring the systems cost of each data representation.
What language is it written in?
Go, according to the project's system overview.
Can it analyse recorded traffic?
Yes. The documentation lists offline usage with pcap files.
Where is the code?
On GitHub, under the traffic-refinery organisation, with a license file in the repository.
What are the best Traffic Refinery alternatives?
Popular alternatives to Traffic Refinery include Network-Security-Based-On-ML, Supervision and NetDeflect. See all Traffic Refinery alternatives




