The repository of paper Brain-on-Switch: Towards Advanced Intelligent Network Data Plane via NN-Driven Traffic Analysis at Line-Speed, published in USENIX NSDI 2024. (The PDF is available at Link)
Our artifact includes the following directories:
- dataset/ contains our datasets and the source code for data pre-processing.
- model/ contains the source code for training our binary RNN model, and converting the trained model into feed forward tables for switch-side implementation.
- PerPacket/ contains the source code for training the tree-based model for handling the flows without dedicated per-flow storage on switch.
- p4/ contains the source code for configuring a Barefoot Tofino 1 (Wedge 100BF-32X) switch and running our RNN-based traffic analysis on it.
- IMIS/ contains the source code for training the Transformer-based model for escalated traffic analysis, and the implementation of our server-side Integrated Model Inference System.
Follow the detailed instructions in each directory to implement the switch-side traffic analysis logic and server-side IMIS.
To forward the escalated traffic between the switch and the server, you need build a network topology to connect them.
Please post a Github issue or send an email to yanjz22@mails.tsinghua.edu.cn if you have any questions.