WindSim

This Repository contains the code used during my internship at the IGE institute, during which my main task was to benchmark the Oceananigans library on different computer architectures.

Organization of the repository

This repository contains four branches:

• main: Original source code - made for parallel
• large_cpu: Adapted for CPU benchmarking
• large_gpu: Adapted for GPU benchmarking
• long-gpu: Adapted for larger simulation on GPU

Each branch contains the following structure

.
├── Notebooks: A collection of jupyter notebooks I used for development
│              (graphs.ipynb usually contains the same figures present in
│              results/visuals)
│
├── results
│   ├── data: A collection of files (in NetCDF format) containing the data 
│   │         obtained from the code. Used to plot the figures.
│   ├── visuals: Collection of figures (graphs) made using the data obtained.
│   ├── logs: Logs from the simulations
│   │   └── time.txt: data obtained from benchmarking Oceananigans
│
└── src:
    ├── format_data.jl
    ├── OceWind.jl: Master script 
    ├── plot_data.jl
    └── simulation: Code used for the simulation
        ├── arch.jl
        ├── grid.jl
        ├── main_sim.jl
        ├── model.jl
        └── simulation.jl

Instructions to run the code

The following need to be installed on your computer first:

• Julia (preferably the latest stable version) [https://julialang.org/downloads/]
• Git [https://git-scm.com/downloads]

Then, run the following commands:

git clone https://github.com/Gaspartcho/OceWind.git
cd OceWind
julia --project=. -e "using Pkg; Pkg.instantiate()"

Once the project has finished pre-compiling, you can run it with the following command:

julia --project=. src/OceWind.jl

You can change parameters some parameters before running the file:

• src/OceWind.jl: change the number of CPU cores used for the simulation
• src/simulation/main_sim.jl: change the size of the grid / the duration simulated.

Warning: the Oceananigans only supports Nvidia GPUs. Trying running the code with an AMD GPU is not guarantied to work.