This presentation was prepared for a demonstration (done on 2020-02-02) of the use of Docker containers with NCAR's ESCOMP/CESM2.2 image on the Google Cloud Shell:

Docker is already available, therefore we can directly pull the latest ESCOMP/CESM-2.2 image from DockerHub:

docker pull escomp/cesm-2.2:latest

and if everything went well this is what you should get:

We can now start the container in detached mode:

docker run -td escomp/cesm-2.2:latest

Take note of the UUID long identifier which starts with 192df0aa73: this string has been randomly generated by the Docker daemon.

To start an interactive process like a bash shell we use docker exec ... bash with the first letters of the UUID (here -t is to allocate a pseudo-tty and -i to keep STDIN open even if not attached):

docker exec -it 192 bash

The promp has now changed to [[email protected] ~]$ to confirm that we are in the container.

We can now, for example, check the version of the operating system inside the container (it is Centos-8 in this case, whereas that was originally Debian in the Google Cloud Shell):

cat /etc/os-release

The container comes with CESM-2.2 and all the software environment necessary to run it, including compilers and libraries.

This particular virtual machine only offers limited resources to effectively run an experiment, therefore we will only try a simple model configuration Moist baroclinic wave with Kessler microphysics and use a coarse resolution:

To create a new case:

create_newcase --case fkessler --compset FKESSLER --res T31_g37 --run-unsupported

Since there are only 2 processors on this machine the number of tasks has to be decreased (the default is 256), and we will run the simulation for 1 day only (instead of 5 days).

To do so we go in the case directory and use the ./xmlchange script:

cd /home/user/fkessler
./xmlchange NTASKS=2
./xmlchange STOP_N=1

Then set-up, build and submit the case:

./case.setup

./case.build

The compilation should take less than 6 minutes and hopefully end with the message "MODEL BUILD HAS FINISHED SUCCESSFULLY".

./case.submit

The missing data is downloaded first, the run begins, finishes after a minute or so and the output files are archived (do not worry about the warning):

The timing information shows that the model throughput was 8.14 simulated_years/day.