Exercise 2: Launching and tracking job steps

1. Ensure that you have a Flux instance running in an allocation and determine the number of cores available to you in that allocation.
2. Download the mpi_hellosleep.c file and build the application with mpicc -o mpi_hellosleep mpi_hellosleep.c.
3. Run mpi_hellosleep as a blocking job step with enough mpi tasks to fill all of the cores available in your Flux instance.
4. Submit two non-blocking job steps that will run mpi_hellosleep with enough mpi tasks to fill all of the cores available in your Flux instance. Verify that only one of the job steps is running.
5. Cancel the running job step. Confirm that the other job step is running and attach to it to monitor its progress.

Notes / Solutions

1. See Exercise 1.
2. On LC systems, mpicc should be in your default path. If it is not, you can run module load intel and module load mvapich2 to get the default mpicc on most LC systems.
3. See ‘Submit blocking job steps’ in Section2.
4. See ‘Submit non-blocking job steps’ in Section2.
5. See ‘Managing job steps’ in Section2.

Section 1 | Section 2 | Exercise 2 | Section 3
Back to index