Flask-based package for monitoring utilisation of GPUs.
Instead of checking each of your machines with nvidia-smi command, a client
running on the machines is sending periodically statistics about GPU
utilisation. The central server is collecting the requests and displays them on
a single website.
Run pip install -r requirements.txt to install all the dependencies.
Most of the settings can be found in config.py.
CLIENT_TIMEOUT allows to set how often a client, i.e. the GPU machine sends
statistics to the central server. You can alter the value with the amount of
seconds you wish to set.
PERMIT_CLIENTS has two fold purpose, first the keys are used to filter out
machines which will be displayed on the website, i.e. we permit only machines
with the specified hostnames, and the corresponding values point to display
names.
Currently the package does not work as a background process so please use
screen, tmux or wsgi.
To run the package as client, i.e. on a GPU machine run
python cli.py -m client
or on the central server execute
python cli.py -m server