ShallowWaterBench

Shallow water benchmarking experiment based of https://github.com/climate-machine/Canary.jl/commit/a58ac057a7224f45bcc450ee533937f7bd06fe17

Setup

julia --project=. -e "using Pkg; Pkg.instantiate(); Pkg.API.precompile()"

Running locally

mpirun -n 4 --oversubscribe julia --project=. original/shallow_water.jl

Running on the supercloud

Setup

Run ./setup.sh everytime that you updated a package, on the login node.

Running normally

run.sc.sh is setup to start 32 mpi ranks per default, using up two full nodes. It is important that you run the setup before starting the job and after the Manifest.toml got updated since the compute nodes don't have access to the internet.

sbatch run.sc.sh

Running the GPU code

run.sc.gpu.sh is setup to use one node with 4 GPUs, in order to pull in the GPU functionality we are using a stacked environment.

sbatch run.sc.gpu.sh