Installation on Maxwell (DESY)

Log into a Maxwell display node

ssh <username>@max-display.desy.de

Create a profile.optimas file in your home directory with the following content

source /etc/profile.d/modules.sh
module purge
module load intel.2020
module load maxwell cuda/11.8
module load mpi/mpich-3.2-x86_64
module load maxwell mamba
. mamba-init

you can do this from the command line by executing cat > ~/profile.optimas and pasting the lines above. To save the file, finalize by pressing Ctrl+D.

Load the source file

source ~/profile.optimas

Create a new environment for optimas

mamba create -n optimas_env python=3.11

Activate the environment

mamba activate optimas_env

Install mpi4py

pip install mpi4py --no-cache-dir

Install optimas with all dependencies if you plan to do Bayesian optimization (see Dependencies for more details).

pip install 'optimas[all]'

Installing FBPIC and Wake-T (optional)

A common use case for optimas is to optimize the output of a plasma acceleration simulation with FBPIC or Wake-T (or both). If you need any of these tools, you can follow the instructions below to install them in your optimas_env environment.

Install FBPIC:

mamba install cudatoolkit=11.8
pip install cupy-cuda11x
pip install fbpic

Install Wake-T:

pip install Wake-T

Install openPMD-viewer for data analysis

pip install openPMD-viewer

Running an optimas job

The following script can be used to submit an optimas job to Maxwell. This example sends the job to the maxgpu queue. It assumes that you need one node with 4 GPUs (NVIDIA A100) and that your optimas script is called run_optimization.py.

#!/bin/bash

#SBATCH --partition=maxgpu
#SBATCH --time=24:00:00
#SBATCH --nodes=1
#SBATCH --constraint="A100&GPUx4"
#SBATCH --job-name  optimas
#SBATCH --output    stdout
#SBATCH --error     stderr
#SBATCH --mail-type BEGIN,END

export SLURM_EXACT=1

# Activate environment
source ~/profile.optimas
mamba activate optimas_env

python run_optimization.py