Talk Running a Coupled General Circulation Model with Python


Running a coupled model can be a daunting task. Prepare an experiment and input files, compile the model, post-process the output, archive it in a way that it's easy to access, and analyze it, and that is only the overview, because each task has many more sublevels. Our development group is quite small, compared to others doing similar research, and the knowledge of how to do each part of the process was scattered between the members. Because of the size of the team we couldn't afford to do repetitive tasks that would consume time needed in other areas, so some degree of automation was necessary too. Besides, new people didn't have a reference of how to do things, or where to find them.

To fix this I began to develop a small tutorial on how to run the model, but after some time I realized that all the steps would fit quite nicely in a script. Fabric fitted perfectly in this idea, because it could be used to run remotely on our center supercomputer and still give freedom to expand it further, like using it into a continuous integration server to monitor test regressions.

Slowly but steadily the script has grown into a suite of commands and made it easier for newcomers to understand how to run the model, at first, and later how to tweak it to their needs. This presentation describes how this suite was implemented, and why it might be a good approach to start a software development workflow in scientific ambients.

tagged by
no related entity