In-network Neural Networks

TL;DR

N2Net demonstrates that commodity network switching chips can efficiently run binary neural networks at high speeds, enabling new in-network AI processing capabilities with minimal hardware modifications.

Contribution

This work introduces N2Net, a system that deploys binary neural networks on network switches, showing potential for high-speed in-network AI with minimal hardware changes.

Findings

Switching chips can process neural networks at billions of packets per second.

Minor modifications could enable more complex neural network models in network devices.

N2Net offers a new building block for future networked AI systems.

Abstract

We present N2Net, a system that implements binary neural networks using commodity switching chips deployed in network switches and routers. Our system shows that these devices can run simple neural network models, whose input is encoded in the network packets' header, at packet processing speeds (billions of packets per second). Furthermore, our experience highlights that switching chips could support even more complex models, provided that some minor and cheap modifications to the chip's design are applied. We believe N2Net provides an interesting building block for future end-to-end networked systems.