dichai/excalibur repository overview

⁠Excalibur

⁠Introduction

This is the docker image for the paper "Efficient Decentralized Federated Singular Vector Decomposition", USENIX ATC'24. This image contains all the necessary dependencies for compiling and running the distributed matrix computations in the paper.

⁠Usage

This image only provides the necessary environments. To reproduce the results or reuse the functions in the paper, please refer to the code in GitHub: https://github.com/Di-Chai/Excalibur⁠. Following we also provide some getting start instructions. Please refer to GitHub for full instructions.

⁠Getting Started Instructions

⁠Step 1: OS and software.

The system is implemented on Linux OS (tested on Ubuntu 20.04) and the following software is needed to reproduce the results:

• Docker
  • Installing instructions.⁠
  • Please also ensure the "docker" command can be executed without sudo.
  • Tested version: Docker version 20.10.21
• Docker-compose
  • Installing instructions.⁠
  • Tested version: docker-compose version 1.28.5
• MiniConda
  • Installing instructions.⁠
  • Tested version: conda 23.5.2

⁠Step 2: Pull the docker image and create the python env.

docker pull dichai/excalibur:v1
docker tag dichai/excalibur:v1 excalibur:v1
conda create -y -n excalibur python=3.8
conda activate excalibur
git clone https://github.com/Di-Chai/Excalibur
cd Excalibur
pip install -r docker/requirements.txt

⁠Step 3: Quick testing.

This will test SVD and its three applications using synthetic data.

conda activate excalibur.
python trial.py test
python python/collect_logs.py logs/
mv logs tmp_test

If everything goes well, you will see the system execution process.

And the tmp_test/result.csv contains all the test results.

Tips: results.csv could be opened using Microsoft Excel (or other tools), which is better for sorting and viewing the results.