Getting started

Setup

The hpcadvisor setup is based on python poetry to handle dependencies and a python virtual environment.

Check poetry installation guidelines HERE. Once poetry is installed run the following commands in your terminal to install and create python virtual environment.

git clone https://github.com/Azure/hpcadvisor.git
cd hpcadvisor
poetry install
poetry shell
cd bin

Generate standalone binary

Alternatively, you can generate a standalone binary file using:

• poetry
• pyinstaller

From project root folder, type:

poetry run pyinstaller src/hpcadvisor/__main__.py  --onefile --name hpcadvisor

---

Simple example

Follow the matrix multiplication example here for detailed instructions.

CLI-based execution

Once the setup above (via poetry) is done, for the matrix multiplication example, you just need to update examples/matrixmult/ui_defaults.yaml with your preferences. For this example, you only need to update subscription, and you should be good to go.

The follow two lines deploy a computing environment specified in ui defaults file and start the data collection (i.e. run the jobs), respectively:

./hpcadvisor deploy create -u ../examples/matrixmult/ui_defaults.yaml
./hpcadvisor collect -n <deploymentname> -u ../examples/matrixmult/ui_defaults.yaml

If you want to only test the plot generator component, skipping the data collection, you need to copy a dataset file to hpcadvisor directory:

mkdir $HOME/.hpcadvisor/
cp ../examples/matrixmult/dataset.json $HOME/.hpcadvisor/

Then request the plot generation:

./hpcadvisor plot -df ../examples/matrixmult/datafilter_matrixmult.yaml

To get the advice (based on pareto-front calculation):

./hpcadvisor advice -df examples/matrixmult/datafilter_matrixmult.yaml

GUI-based execution

One can use the browser version and click the buttons for the different operations. To pre-fill user input, specify the input file as showed below:

./hpcadvisor gui -u ../examples/matrixmult/ui_defaults.yaml

---