2. Installation and basic usageΒΆ

Prerequisite

Comprehensive calculation utility moller included in HTP-tools requires the following programs and libraries:

  • Python 3.x

  • ruamel.yaml module

  • tabulate module

  • GNU Parallel (It must be installed on servers or compute nodes on which the job script is executed.)

Official pages

Downloads

moller can be downloaded by the following command with git:

$ git clone https://github.com/issp-center-dev/Moller.git

Installation

Once the source files are obtained, you can install moller by running the following command. The required libraries will also be installed automatically at the same time.

$ cd ./Moller
$ python3 -m pip install .

The executable files moller and moller_status will be installed.

Directory structure

.
|-- LICENSE
|-- README.md
|-- pyproject.toml
|-- docs/
|   |-- ja/
|   |-- en/
|   |-- tutorial/
|-- src/
|   |-- moller/
|       |-- __init__.py
|       |-- main.py
|       |-- platform/
|       |   |-- __init__.py
|       |   |-- base.py
|       |   |-- base_slurm.py
|       |   |-- base_pbs.py
|       |   |-- base_default.py
|       |   |-- ohtaka.py
|       |   |-- kugui.py
|       |   |-- pbs.py
|       |   |-- default.py
|       |   |-- function.py
|       |   |-- utils.py
|       |-- moller_status.py
|-- sample/

Basic usage

moller is a tool to generate batch job scripts for supercomputers in which programs are run in parallel for a set of execution conditions using concurrent execution features.

  1. Prepare job description file

    First, you need to create a job description file in YAML format that describes the tasks to be executed on supercomputers. The details of the format will be given in File Format section of the manual.

  2. Run command

    Run moller program with the job description file, and a batch job script will be generated.

    $ moller -o job.sh input.yaml

  3. Run batch jobs

    Transfer the generated batch job scripts to the supercomputer. Prepare a directory for each parameter set, and create a list of the directory names in a file list.dat. Note that the list contains the relative paths to the directory where the batch job is executed, or the absolute paths.

    Once the list file is ready, you may submit a batch job. The actual command depends on the system.

    • In case of ISSP system B (ohtaka)

      In ohtaka, slurm is used for the job scheduling system. In order to submit a batch job, a command sbatch is invoked with the job script as an argument. Parameters can be passed to the script as additional arguments; the name of list file is specified as a parameter.

      $ sbatch job.sh list.dat

      If the list file is not specified, list.dat is used by default.

    • In case of ISSP system C (kugui)

      In kugui, PBS is used for the job scheduling system. In order to submit a batch job, a command qsub is invoked with the job script. There is no way to pass parameters to the script, and thus the name of the list file is fixed to list.dat.

      $ qsub job.sh

  4. Check the status of the calculation

    After the job finishes, you may run the following command

    $ moller_status input.yaml list.dat

    to obtain a report whether the calculation for each parameter set has been completed successfully.

  5. Retry/resume job

    In case the job is terminated during the execution, the job may be resumed by submitting the batch job again with the same list file. The yet unexecuted jobs (as well as the unfinished jobs) will be run.

    • In case of ISSP system B (ohtaka)

    $ sbatch job.sh list.dat

    To retry the failed tasks, the batch job is submitted with --retry command line option.

    $ sbatch job.sh --retry list.dat

    • In case of ISSP system C (kugui)

    For kugui, to retry the failed tasks, the batch job script should be edited so that retry=0 is changed to be retry=1.

    $ qsub job.sh

    Then, the batch job is submitted as above.

References

[1] O. Tange, GNU Parallel - The command-Line Power Tool, ;login: The USENIX Magazine, February 2011:42-47.