SimVascular and SvSolver

These are Docker builds of SimVascular and SvSolver intended to be run with Singularity. It clones and builds simvascular from the SimVascular repository master branch, and SvSolver from it's respective repository. If you intend to use this to build a container for research, you should customize the Dockerfile build to install a release.

Note The Dockerfiles here are redundant in having one per openmpi version - we will want to make a build recipe with a grid of builds that provide the bases and MPI flavors and versions that are most commonly needed as build args.

The intended use case is the following:

SimVascular

• SimVascular with OpenMPI version 1.10.7 tags latest and 2020-04 built from Dockerfile.simvascular

Build

Clone the repository

git clone https://github.com/vsoch/simvasvular
cd simvascular

Build the container, and tag:

docker build -t vanessa/simvascular .
docker tag vanessa/simvascular:latest vanessa/simvascular:2020-04

You can then pull directly from your Docker daemon:

singularity pull docker-daemon://vanessa/simvascular:2020-04

You can transfer (e.g., with scp) the container, or even better, if you want you can push to some Docker Registry to share or pull from another host (e.g., your cluster):

# pushes all tags
docker push vanessa/simvascular
singularity pull docker://vanessa/simvascular:2020-04

Usage

Singularity

The easiest thing to do is then run via Singularity. You can pull the container first (see above) and then run the container:

singularity run simvascular_2020-04.sif

By default, the container entrypoint is to the executable simvascular that will open up the interface:

If you want to use the solver, do this:

singularity exec simvascular svsolver

The process ID for myrank (0) is (23395).


The number of processes is 1.

Solver Input Files listed as below:
------------------------------------
 Local Config: solver.inp 
 Input file does not exist or is empty or perhaps you forgot mpirun?

or the svfsi solver (not sure about the difference) - note that you need mpi 1.10 on the host for this to work. You probably don't have it (it's not on Sherlock). Sorry. See this issue to see why it's needed and there's nothing I can do about it.

singularity exec simvascular svfsi

Note that if you are on sherlock, you can always load additional libraries that you need, a simple search with spider looks like this:

module spider mpi

svSolver

All containers are on Docker Hub under vanessa/svsolver. Associated Dockerfiles and tags include:

• Dockerfile.svsolver builds svsolver with OpenMPI version 1.10.7, tags latest and 2020-04
• Dockerfile.svsolver-openmpi-3.1.4 builds svsolver with OpenMPI version 3.1.4, tags openmpi-3.1.14

These Dockerfiles are reudundant and should be replaced with build args to determine versions.

Build

You can again build the svSolver from the svSolver Dockerfile.

docker build -f Dockerfile.svsolver -t vanessa/svsolver .
docker tag vanessa/svsolver:latest vanessa/svsolver:2020-04

And pull with Singularity from your docker daemon.

singularity pull docker-daemon://vanessa/svsolver

You can also push to a Docker registry to pull to some host:

docker push vanessa/svsolver
singularity pull docker://vanessa/svsolver:2020-04

Usage

Docker

You can try running the Docker container, but it likely isn't so useful without MPI:

$ docker run -it vanessa/svsolver

The process ID for myrank (0) is (1).


The number of processes is 1.

Solver Input Files listed as below:
------------------------------------
 Local Config: solver.inp 
 Input file does not exist or is empty or perhaps you forgot mpirun? 

Singularity

For Singularity, you can first pull the image from your remote registry:

singularity pull docker://vanessa/svsolver:2020-04

The entrypoint is to binaries in /code/svSolver/BuildWithMake/Bin, which are intended for linux but have .exe extensions:

$ docker run -it --entrypoint bash vanessa/svsolver
root@47d1dee35d17:/code# ls svSolver/BuildWithMake/Bin/
svpost-gcc-gfortran.exe  svpost.exe  svpre-gcc-gfortran.exe  svpre.exe  svsolver-gcc-gfortran-nompi.exe  svsolver-nompi.exe

The current entrypoint is "svsolver-nompi.exe," indicating that there is no MPI, however ideally there would be automated builds providing several versions of MPI inside the container.