build-arm

Build Arm

To generate the arm images we will build on an AWS VM. Note that originally I was using one arm variant, but quickly learned that compatiility (on the level of the architecture) was not guaranteed and wanted to extend the matrix. So here is how to create a base arm image. An early experiment to see if QEMU would have distinct archtecture information is in experiment.md. Note that it does not - it reflects the host architecture. Note that we will use some of the commands there for setup (and not repeat them here). You should start this small tutorial with a running VM that has docker installed.

Build

Shell into the instance with your PEM:

ssh -o 'IdentitiesOnly yes' -i path-to-key.pem ec2-user@ec2-52-55-54-40.compute-1.amazonaws.com

Install oras (for arm)!:

VERSION="1.1.0"
curl -LO "https://github.com/oras-project/oras/releases/download/v${VERSION}/oras_${VERSION}_linux_arm64.tar.gz"
mkdir -p oras-install/
tar -zxf oras_${VERSION}_*.tar.gz -C oras-install/
sudo mv oras-install/oras /usr/local/bin/
rm -rf oras_${VERSION}_*.tar.gz oras-install/

Note that we aren't going to use bulidx here. Clone this repository for the Dockerfiles.

git clone https://github.com/rse-ops/lammps-matrix
cd lammps-matrix

Let's now build our images and artifacts. Since we get no benefit from building inside the container, let's run it post build. We will use generate.sh. You'll need to login to ghcr.io with a GitHub personal access token for password.

docker login ghcr.io

And then run the builds and extractions and pushes. Note that we are manually specifying the hasGpu command, and we are sticking with building arm on arm. We will do another instance for x86. Note that we load first so we have the layers.

# create directory for output specs
cd build-arm
mkdir -p ./specs

# These are gpu, openmpi and ubuntu
hasGpu=yes
for version in 20.04 22.04; do
    tag=openmpi-ubuntu-gpu-${version}-arm64
    image=ghcr.io/rse-ops/lammps-matrix:${tag}
    echo "Building $image"
    docker build --build-arg tag=${version} --network=host --tag $image ../openmpi-ubuntu-gpu/
    docker push ${image}
    cmd=". /etc/profile && /tmp/data/generate.sh $hasGpu /tmp/data/specs/compspec-${tag}.json"
    docker run --entrypoint bash -v $PWD:/tmp/data -it ${image} -c "$cmd"
    oras push ${image}-compspec --artifact-type application/org.supercontainers.compspec ./specs/compspec-${tag}.json:application/org.supercontainers.compspec
done


# gpu, openmpi rocky (hpc-tools for intel-mpi does not support arm)
for version in 8 9; do
    tag=openmpi-rocky-gpu-${version}-arm64
    image=ghcr.io/rse-ops/lammps-matrix:${tag}
    echo "Building $image"
    docker build --build-arg tag=${version} --network=host --tag $image ../openmpi-rocky-gpu/
    docker push ${image}
    cmd=". /etc/profile && /tmp/data/generate.sh $hasGpu /tmp/data/specs/compspec-${tag}.json"
    docker run --entrypoint bash -v $PWD:/tmp/data -it ${image} -c "$cmd"
    oras push ${image}-compspec --artifact-type application/org.supercontainers.compspec ./specs/compspec-${tag}.json:application/org.supercontainers.compspec
done

hasGpu=no
for version in 20.04 22.04; do
    for mpi in openmpi mpich; do
        tag=${mpi}-ubuntu-${version}-arm64
        image=ghcr.io/rse-ops/lammps-matrix:${tag}
        echo "Building $image"
        docker build --build-arg tag=${version} --network=host --tag $image ../${mpi}-ubuntu/
        docker push ${image}
        cmd=". /etc/profile && /tmp/data/generate.sh $hasGpu /tmp/data/specs/compspec-${tag}.json"
        docker run --entrypoint bash -v $PWD:/tmp/data -it ${image} -c "$cmd"
        oras push ${image}-compspec --artifact-type application/org.supercontainers.compspec ./specs/compspec-${tag}.json:application/org.supercontainers.compspec
    done
done

Builds

Here are the final builds (with artifacts each) for the above.

Image	Architecture	OS	OS Version	MPI	MPI version	GPU
openmpi-ubuntu	linux/arm64	ubuntu	20.04	openmpi		no
openmpi-ubuntu	linux/arm64	ubuntu	22.04	openmpi		no
openmpi-ubuntu-gpu	linux/arm64	ubuntu	20.04	openmpi		yes
openmpi-ubuntu-gpu	linux/arm64	ubuntu	22.04	openmpi		yes
openmpi-rocky-gpu	linux/arm64	rocky	8	openmpi		yes
openmpi-rocky-gpu	linux/arm64	rocky	9	openmpi		yes
mpich-ubuntu	linux/arm64	ubuntu	20.04	mpich		no
mpich-ubuntu	linux/arm64	ubuntu	22.04	mpich		no

These will be represented in manifests.yaml