Singularity is the preferred method for running containers on HPC systems. For laptop and desktop Linux machines, it is often a more user friendly way of running containers than Docker. Singularity can use existing Docker images.
On Debian and Ubuntu, Singularity is available in the default
repository in the singularity-container package. So installing
Singularity should be as simple as running
sudo apt-get install singularity-container
On Redhat and Centos, Singularity is available through EPEL. The Singularity web site has detailed installation instructions.
On HPC systems, if Singularity is available, it is probably installed as a module. Run
module av
and look for singularity.
Once Singularity is installed, the next step is to download and convert the image
singularity pull docker://wlandry/sdpb:2.5.1
Depending on your version of Singularity, this will create a file
named sdpb-2.5.1.simg or sdpb_2.5.1.sif.
Singularity should automatically mount your home directory. If your
Singularity image is named sdpb-2.5.1.simg, you can invoke the SDPB
programs by prepending the command with
singularity exec sdpb-2.5.1.simg
So to convert the XML input file at /home/user/input.xml, run the command
singularity exec sdpb-2.5.1.simg mpirun -n 4 pvm2sdp 1024 /home/user/input.xml /home/user/input
This uses 4 cores when running pvm2sdp. You can change that number to match your own machine.
To find a primal-dual solution,
singularity exec sdpb-2.5.1.simg mpirun -n 4 sdpb --precision=1024 --procsPerNode=4 -s /home/user/input
In theory, Singularity can be used to run jobs across multiple nodes. We have not been able to make that work yet. So for large, multi-node jobs, it is still recommended to build from source using the instructions in Install.md.