Globus Compute Golf Simulation Demo

This demo showcases executing simulations at scale using Globus Compute.

As a proxy for a real scientific simulation, this demo simulates dropping golf balls on a green. A golf green is randomly generated using Perlin noise, and the simulation is performed using PyBullet.

• Installation
• Visualize a Simulation
• Simulate at Scale
• Contributing

Installation

1. Clone the repository and navigate to the project directory:

   git clone https://github.com/globus-labs/globus-compute-golf-demo
   cd globus-compute-golf-demo

2. Create a virtual environment: This demo requires Python 3.9 or later.

   python -m venv venv
   . venv/bin/activate

   You may use alternate environment managers, such as Conda, provided you have pip installed. We do not recommend installing the packages into your system Python.
3. Install the package and dependencies:

   pip install .

4. Configure a Globus Compute Endpoint: This step can be performed on your local machine or on a remote machine. You will need to authenticate with Globus Auth if you have not used Globus Compute on your local device before. A UUID for your endpoint will be returned; save this for later (you can find it later with globus-compute-endpoint list).

   globus-compute-endpoint configure golf-demo
   globus-compute-endpoint start golf-demo

   For more information on configuring Endpoints, check out the docs.

Visualize a Simulation

We will start with running a single simulation locally to visualize what the simulation looks like. This will generate a random terrain and drop a ball at a random position.

python run_single.py --num-balls 100

You should see a pop up window that looks like the following.

The parameters of the simulation are fully configurable. Use python run_single.py --help to see all of the options.

For example, set --width 50 to increase the size of the terrain. This will also zoom the camera out so it may be difficult to see the balls; the ball size can be changed with --ball-diameter.

Simulate at Scale

Next, we can run the same simulation at scale by generating n random initial ball positions and submitting one simulation per ball to Globus Compute. Globus Compute will manage the parallel execution of tasks on the Endpoint you configured. Replace <UUID> with the UUID of your Endpoint.

python run_globus_compute.py --num-balls 1000 --endpoint <UUID>

Tip

Passing --process-pool 8 instead of --endpoint <UUID> will run the simulations across a pool of 8 local processes rather than via Globus Compute.

Warning

Increasing the width of the terrain or the resolution of the terrain will increase the size of the mesh map. Globus Compute has a 10MB payload limit which can be exceeded if the terrain is too large.

Because the simulations are executed in parallel on remote processes, the simulations are performed in a "headless" mode (i.e., the simulation is not rendered to the screen). The script visualizes the results of the simulations via plots that are saved to images/.

2D Contour Plot (images/contour.png)

3D Terrain Heatmap (images/terrain.png)

Contributing

To contributed to this repository:

1. Fork and/or create a branch for your features/changes. The main branch of this repository has branch protection enabled so all changes must be made via a pull request.
2. Install the package with development dependencies.

   pip install -e .[dev]

3. Make your changes.
4. Run the pre-commit checks and fix any issues.

   pre-commit run --all-files

5. Commit your changes and open a pull request.