Overview

This repository contains the source code to reproduce the experiments of the paper:

• Leaky DNN: Stealing Deep-learning Model Secret with GPU Context-switching Side-channel (DSN 2020)

Tested Setup

• Our study is carried out on a workstation equipped with Nvidia GeForce GTX 1080 TI.
• The graphics driver version is 384.11.
• The CUDA version is V9.0.176 and the cuDNN version is 7.4.2.
• The workstation has installed Ubuntu 16.04 and we use Tensorflow 1.12.0.

Extract GPU performance counters

• To run it, you just need to type sh run_spykernel.sh.
• To add more performance counters, you just need to add the names of metrics in CUPTI_Conv_Metrics.cu or event in CUPTI_Conv_Event.cu.
• The GPU performance counters will be stored into conv_metrics.csv and conv_event.csv.

Citation

If you make any use of this code for academic purposes, please cite the paper:

@inproceedings{wei2020leaky,
  title={Leaky dnn: Stealing deep-learning model secret with gpu context-switching side-channel},
  author={Wei, Junyi and Zhang, Yicheng and Zhou, Zhe and Li, Zhou and Al Faruque, Mohammad Abdullah},
  booktitle={2020 50th Annual IEEE/IFIP International Conference on Dependable Systems and Networks (DSN)},
  pages={125--137},
  year={2020},
  organization={IEEE}
}