ChainNet

ChainNet is a customized graph neural network model designed to evaluate the reliability of edge AI deployments. An edge AI service consists of a chain of deep neural network (DNN) fragments. For a given set of service chains, deployment refers to allocating their fragments across available devices.

The input to ChainNet is a deployment plan characterized by a set of files covering two main aspects: (1) the mapping relationship between fragments and devices, and (2) the features of services, fragments, and devices. The output includes the throughput and latency of each service chain.

Quick start

1. Clone repo:

git clone https://github.com/imperial-qore/ChainNet.git
cd ChainNet/

2. Create an environment in line with the environment.yaml file:

conda env create -f environment.yaml && conda activate ChainNet_env

or install the packages in a Python 3.8 environment:

python -m venv env && source env/bin/activate && pip install -r requirements.txt

3. To train and test ChainNet, use the following scripts:

python main.py

python evaluation.py

Cite this work

Our work is accepted by the 54th Annual IEEE/IFIP International Conference on Dependable Systems and Networks (DSN). Cite our work using the bibtex entry below.

@inproceedings{chainnet2024dsn,
  title={ChainNet: A Customized Graph Neural Network Model for Loss-aware Edge AI Service Deployment},
  author={Niu, Zifeng and Roveri, Manuel and Casale, Giuliano},
  booktitle={IEEE/IFIP International Conference on Dependable Systems and Networks (DSN)},
  year={2024},
  organization={IEEE}
}

License

BSD-3-Clause. Copyright (c) 2024, Zifeng Niu. All rights reserved.

See the license file for more details.