Cheetah: Lean and Fast Secure Two-Party Deep Neural Network Inference

This repo contains a proof-of-concept implementation for our Cheetah paper. The codes are still under heavy developments, and should not be used in any security sensitive product.

Cheetah -> Secure Processing Unit

Most of the Cheetah protocols has been re-written in the SecretFlow project. Check the code here.

Q&A (Updating)

See QA.md.

Repo Directory Description

• include/ Contains implementation of Cheetah's linear protocols.
• SCI/ A fork of CryptFlow2's SCI library and contains implementation of Cheetah's non-linear protocols.
• networks/ Auto-generated cpp programs that evaluate some neural networks.
• pretrained/ Pretrained neural networks and inputs.
• patch/ Patches applied to the dependent libraries.
• credits/ Licenses of the dependencies.
• scripts/ Helper scripts used to build the programs in this repo.

Requirements

• openssl
• c++ compiler (>= 8.0 for the better performance on AVX512)
• cmake >= 3.13
• git
• make
• OpenMP (optional, only needed by CryptFlow2 for multi-threading)

Building Dependencies

• Run bash scripts/build-deps.sh which will build the following dependencies

  • emp-tool We follow the implementation in SCI that using emp-tool for network io and pseudo random generator.
  • emp-ot We use Ferret in emp-ot as our VOLE-style OT.
  • Eigen We use Eigen for tensor operations.
  • SEAL We use SEAL's implementation for the BFV homomorphic encryption scheme.
  • zstd We use zstd for compressing the ciphertext in SEAL which can be replaced by any other compression library.
  • hexl We need hexl's AVX512 acceleration for achieving the reported numbers in our paper.

• The generated objects are placed in the build/deps/ folder.

• Build has passed on the following setting

  • MacOS 11.6 with clang 13.0.0, Intel Core i5, cmake 3.22.1
  • Red Hat 7.2.0 with gcc 7.2.1, Intel(R) Xeon(R), cmake 3.12.0
  • Ubuntu 18.04 with gcc 7.5.0 Intel(R) Xeon(R), cmake 3.13
  • Ubuntu 20.04 with gcc 9.4.0 Intel(R) Xeon(R), cmake 3.16.3

Building Cheetah and SCI-HE Demo

• Run bash scripts/build.sh which will build 6 executables in the build/bin folder
  • resnet50-cheetah
  • sqnet-cheetah
  • densenet121-cheetah
  • resnet50-SCI_HE
  • sqnet-SCI_HE
  • densenet121-SCI_HE

Local Demo

1. On one terminal run bash scripts/run-server.sh cheetah sqnet. The program will load the pretrained model in the folder pretrained/ which might takes some time when the pretrained model is huge.

2. On other terminal run bash scripts/run-client.sh cheetah sqnet. The program will load the prepared input image in the folder pretrained.

   • replace cheetah by SCI_HE to execute the CryptFlow2's counterpart.
   • replace sqnet by resnet50 to run on the ResNet50 model.

You can change the SERVER_IP and SERVER_PORT defined in the scripts/common.sh to run the demo remotely. Also, you can use our throttle script to mimic a remote network condition within one Linux machine, see below.

Mimic an WAN setting within LAN on Linux

• To use the throttle script under scripts/throttle.sh to limit the network speed and ping latency (require sudo)
• For example, run sudo scripts/throttle.sh wan on a Linux OS which will limit the local-loop interface to about 400Mbps bandwidth and 40ms ping latency. You can check the ping latency by just ping 127.0.0.1. The bandwidth can be check using extra iperf command.