Singularity Examples
Introduction
Here we show some quick examples on how singularity can be used. If you are unsure about the details please read the general description of Singularity or the singularity documentation.
Building and Running a Container with GPU support
The following recipe has been written to work with the CUDA-Toolkit and is based on an Ubuntu distribution environment:
Bootstrap: docker
From: ubuntu:18.04
%post
apt-get -y update
apt-get -y install python3-numpy
apt-get -y install python3-sympy
apt-get -y install python3-pip
pip3 install sympy==1.3
apt-get -y install python3-matplotlib
apt-get -y install ffmpeg
apt-get -y install --no-install-recommends --no-install-suggests nvidia-cuda-toolkit
mkdir /metacode
mkdir /scripts
mkdir /staticcode
mkdir /working
mkdir /output
%environment
%runscript
umask go-rwx
To make use of it you have to build it on your own local linux system and then transfer it to your cluster directory:
- Create a copy of the script shown above to a file:
vim cuda9python3.def # insert the content and save&quit the file
- Build a singularity image from the .def file (You will need ~ 1.3 GB free disk space for this image):
sudo singularity build cuda9python3.simg cuda9python3.def
- Copy the image file to the cluster
rsync cuda9python3.simg abcd1234@carl.hpc.uni-oldenburg.de:~/<desired_path>
- After logging in, load the modules for Singularity and cuda:
module load hpc-env/6.4
module load Singularity
module load CUDA-Toolkit
- Navigate to the image folder and start using it:
cd ~/path/to/cuda9python3.simg
singularity exec --nv <your_commands_here>
Please notice, that using CUDA-Toolkit requires logging in to a GPU node. That means, that the commands on the cluster shown above should be written to a batch script and transferred to a GPU partition (e.g. mpcg.p).
Also keep in mind, that the singularity option --nv is mandatory if you want to use the GPU functionality. With it, the container can make use of the clusters GPU driver libraries.
