Overview / Usage

The purpose of this project is to encourage others to start building Network AI solutions by lowering the cost of entry and provide a framework for AI to be introduced in the SDN data plane.

Current goals of this project:

• Build, collect and share samples of network traffic
• Build tools to convert, annotate and normalized captured networking samples for use in neural networks
• Build an initial set of models designed to demonstrate AI on networking, to be shared publicly
• Provide a mechanism for others to share network traffic samples or AI models

Methodology / Approach

• Standard Linux tooling will be used to capture sample networking traffic. Tools will be built to clean, annotate and normalize data.

• AI models will be built using open source tools such as TensorFlow or PyTorch, depending on the problem.

• Data and models to be shared publicly on an existing exchange. If no exchange exists that is suitable, then models and samples will be shared on GitHub.

Technologies Used

• Linux
• tcpdump
• Python (for training)
• NVIDIA GPUs for training
• ML Framework TBD (Likely TensorFlow)
• Movidius (to demonstrate AI hardware offload in decision)