Workflow

ULHPC Workflow

Your typical journey on the ULHPC facility is illustrated in the below figure.

Typical workflow on UL HPC resources

You daily interaction with the ULHPC facility includes the following actions:

Preliminary setup

1. Connect to the access/login servers
   • This can be done either by ssh (recommended) or via the ULHPC OOD portal
   • (advanced users) at this point, you probably want to create (or reattach) to a screen or tmux session
2. Synchronize you code and/or transfer your input data using rsync/svn/git typically
   • recall that the different storage filesystems are shared (via a high-speed interconnect network) among the computational resources of the ULHPC facilities. In particular, it is sufficient to exchange data with the access servers to make them available on the clusters
3. Reserve a few interactive resources with salloc -p interactive [...]
   • recall that the module command (used to load the ULHPC User software) is only available on the compute nodes
   • (eventually) build your program, typically using gcc/icc/mpicc/nvcc..
   • Test your workflow / HPC analysis on a small size problem (srun/python/sh...)
   • Prepare a launcher script <launcher>.{sh|py}

Then you can proceed with your Real Experiments:

1. Reserve passive resources: sbatch [...] <launcher>
2. Grab the results and (eventually) transfer back your output results using rsync/svn/git

---

Last update: December 2, 2024