Table of contents

• Docker
  • Build container
    • Test
    • Notes
  • Compile and run modelE
    • Enter container
    • Compile
    • Run
  • Deploy for sharing
• Singularity
  • Use on discover

Docker

• Instructions on using Docker to run ModelE
• The Docker container contains all the tools needed to compile and run ModelE such as gfortran, gdb, netcdf, etc.
• ModelE source code and ModelE_Support (input files) are assumed to be on the host computer, and accessed via mount points.

Build container

docker build -t modele .

Test

• Replace $(pwd)/modelE with your /path/to/modelE source code directory.
• Replace ${HOME}/data/ModelE_Support with your /path/to/Model_Support.

# run it with default CMD
docker run -it modele

# run interactively
docker run -it modele bash
exit

#       --mount type=bind,src=${HOME},dst=/home/user \
docker run -it --user $(id -u):$(id -g) \
       --mount type=bind,src=$(pwd)/../modelE,dst=/modelE \
       --mount type=bind,src=${HOME}/data/ModelE_Support,dst=/ModelE_Support \
       modele bash
ls -l /modelE
ls -l /ModelE_Support
echo $MODELERC
ls -l $MODELERC
exit

Notes

• modelE source code
  • Editing of source files is assumed to occur on the host computer.
  • You can mount your host modelE source code folder anywhere in the container.
  • In the example above, the source is mounted at /modelE.
  • In the example above, we mounted the host folder (found in the current folder at $(pwd)/modelE) to the container /modelE with the line --mount type=bind,src=$(pwd)/modelE,dst=/modelE.
• ModelE_Support
  • You can mount your host ModelE_Support folder anywhere in the container.
  • In this example, ModelE_Support is mounted at /ModelE_Support.
  • If you mount it anywhere else, you need to edit the ${MODELERC} file.

Compile and run modelE

Enter container

docker run -it \
       --mount type=bind,src=$(pwd)/modelE,dst=/modelE \
       --mount type=bind,src=${HOME}/data/ModelE_Support,dst=/ModelE_Support \
       modele \
       bash

Compile

cd /modelE/decks/

# 8x10 lat x lon; no ocean; model month takes O(minute)
RUNNAME=docker_test01
make rundeck RUNSRC=nonProduction/nonProduction_E_AR5_C12 RUN=${RUNNAME}
# fetch input files for this run
../exec/get_input_data -w ${RUNNAME} /ModelE_Support/prod_input_files/

# # 4x5 lat x lon; ocean; model month takes 10x longer
# RUNNAME=docker_test02
# make rundeck RUNSRC=E1oM20 RUN=${RUNNAME}

make clean RUN=${RUNNAME}

# compile for debugging
make -j setup RUN=${RUNNAME} COMPILE_WITH_TRAPS=YES EXTRA_FFLAGS="-g -O0 -fallow-argument-mismatch -I/usr/include/x86_64-linux-gnu " EXTRA_LFLAGS="-g -O0" | tee compile.out

# compile for fast run
make -j setup RUN=${RUNNAME} COMPILE_WITH_TRAPS=NO EXTRA_FFLAGS="-O2 -fallow-argument-mismatch -I/usr/include/x86_64-linux-gnu " EXTRA_LFLAGS="-O2" | tee compile.out

Run

../exec/runE ${RUNNAME} -cold-restart -np 4
../exec/runE ${RUNNAME} -np 4

Deploy for sharing

# docker tag local-image:tagname new-repo:tagname
docker tag modele mankoff/modele:gfortran
docker login -u "user" -p "pass" docker.io
docker push mankoff/modele:gfortran

• https://hub.docker.com/r/mankoff/modele

Singularity

This section not yet completed.

• https://www.nccs.nasa.gov/nccs-users/instructional/containers

Use on discover

• You cannot run docker on discover, but you can run docker images using singularity.
• singularity is like docker but for HPC systems

Pulling down VMs can take some space, don’t do it in ~/

cd ${NOBACKUP}
export SINGULARITY_CACHEDIR="$NOBACKUP/.singularity"
export SINGULARITY_TMPDIR="$NOBACKUP/.singularity"

module load singularity
# singularity build modele.sif docker-daemon://modele # local
# singularity pull lynx.sif docker://nbrown/lynx
# singularity exec -B ./:${TMPDIR} lynx.sif lynx http://www.giss.nasa.gov