2. Environment

In this section, we will help you building the environment.

2.1. Software Dependencies

2.1.1. System Requirements

Currently, this system only support Linux platforms.

Recommended memory sizes for different dataset and models are shown in the following table. Increasing the swap space is a good solution if your compute does not have enough memory.

Dataset	ML Model	Required RAM size
MNIST	MLP	32 GB / 100 Clients
MNIST	LeNet	80 GB / 100 Clients
FEMNIST	LeNet	60 GB / 100 Clients
CelebA	LeNet	128 GB / 100 Clients
Sent140	LSTM	128 GB / 100 Clients

2.1.2. Docker & Docker Compose (without sudo)

Docker is an indispensable component to virtualize the software and network environment in order to isolate each instance. Follow these instructions to install Docker on your machine. Follow the guideline on Link to manage docker as non-root user. Docker has been successfully installed if you got a similar output like this:

$ docker --version
Docker version 20.10.2, build 2291f61

Docker Compose is a tool aimed at easing multi-container operations. Install it under the guidance and check it with:

$ docker-compose version
docker-compose version 1.27.4, build 40524192
docker-py version: 4.3.1
CPython version: 3.7.4
OpenSSL version: OpenSSL 1.1.1c  28 May 2019

2.1.3. Python

Though most of the programs can be boosted inside containers, it is still recommended to have a Python interpreter installed on your machine for the convenience and immediacy of type checking, auto complete, and so on during code profiling or some advanced, self-defined operations.

For the sake of compatibility with main components of the system, the version of the interpreter should be in the range of 3.5–3.8 currently. You can find how to install them here.

2.1.4. Git (Optional)

Git is a distributed version control system. We use it and host this project on Github. Feel free to skip this because you also can download the project as a zip file from the web page.

Clone the project from Github in this way if have installed git.

$ git clone https://github.com/Di-Chai/FedEval.git -b dev-0.1

2.2. Build the Docker Image

The image we build will be used by each instance (both server and clients).

First, clone the project from GitHub:

git clone https://github.com/Di-Chai/FedEval.git

Enter the project directory and build the image.

cd docker
sudo docker build . -t fedeval:v1

Check image status. This command will list all docker images on your machine.

sudo docker image ls

And you’ll got something like these if nothing goes wrong. It’s okay if the image ID is different.

REPOSITORY     TAG     IMAGE ID       CREATED         SIZE
fedeval        v1      2386be014ab5   3 months ago    4.39GB

2.3. Download the dataset

The dataset is available at : resources

Download the data.tar.gz, and put it at the project directory. Using the following commend to unpack the data:

mv data.tar.gz FedEval
cd FedEval
tar -zxvf data.tar.gz