To monitor RAM and GPU usage on multiple servers.

In a computer science lab or company, you usually have multiple servers and GPUs running many deep learning experiments. You want to know which device is working and which is available at a glance with a minimum setup.

Screenshots

Installation

git clone https://github.com/kehanlu/server-monitor
cd server-monitor
pip install -r requirements.txt

• nvidia-smi: https://www.nvidia.com

Usage

Server

"Server" means the server you want to monitor.

1. Go to server you want to monitor

   • You have to be sure that nvidia-smi command is installed.

2. run the command to start an API.

gunicorn -w 1 -k uvicorn.workers.UvicornWorker -b 0.0.0.0:23333 server:app --daemon

Master

"Master" means the web server which is going to fetch data from each servers. You can run this web server on any computer. In some case, you might want this web server to be accessible from public network, but still hide servers behind a firewall.

1. create a file named config.py

2. In config.py, you need to have a list of server ips. Then the web server will iterate from the list and GET the API at http://{ip}:23333.

• server_ips
• site_title(optional): the title of website
• top_message(optional): the message shows on the top

CONFIG = {
    "site_title": "Server status",
    "top_message": "Hello world",
    "server_ips": [
        "192.168.0.2",
        "192.168.0.3",
        "192.168.0.4",
    ],
}

3. run the command to start the server.

gunicorn -w 1 -b 0.0.0.0:8787 master:app

4. Visit 127.0.0.1:8787 or <your_ip>:8787 to see the website.

Contribution

Pull requests are welcome. This is still an early project (and just for fun).

TODOs:

• Fast installation script.
• Handle error.
• Use Nginx to serve the sites.
• Use CI/CD to automatically update projects on servers.