Execution Guide

Execution Guide#

Summary: Choosing the Right Execution Strategy

Selecting the appropriate execution mode and parallel strategy depends on the scale of your model, the type of study you’re conducting, and your computing environment.

Quick Reference Table#

Use Case

Mode

Example Command

Notes

Exploring commands

Interactive

OpenSees or python

Best in terminal or small scripts

Full analysis run (sequential)

Non-Interactive

OpenSees model.tcl

Most common for local or small batch jobs

Batch run with varying inputs

Arg-based Non-Interactive

OpenSees model.tcl 0.05 10 input.at2

Great for sweeps, job arrays

Parallel simulation (Tcl)

MPI via OpenSeesMP/SP

mpiexec -np 4 OpenSeesMP model.tcl

Choose MP or SP based on domain decomposition style

Parallel simulation (Python)

MPI via OpenSeesPy

mpiexec -np 4 python model.py

Supports OpenSeesMP-style decomposition internally

Independent jobs (Python)

concurrent.futures

ProcessPoolExecutor(…).map(…)

For embarrassingly parallel jobs without communication

Decision Flow#

  1. Do you need multiple processes or distributed memory?

    • ✅ Yes → Use MPI with OpenSeesMP or OpenSeesPy

    • ❌ No → Continue

  2. Do you want to vary inputs across runs?

    • ✅ Yes → Use command-line arguments and loop/array strategy

    • ❌ No → Run script directly (non-interactive)

  3. Are jobs independent (no MPI communication)?

    • ✅ Yes → Use concurrent.futures in Python for batch submission

    • ❌ No → Stick with mpiexec and an MPI-based script

  4. Running in DesignSafe Jupyter?

    • Use terminal to launch parallel jobs

    • Submit to Tapis or HPC cluster for large-scale runs

With these strategies and tools, you can scale from small test scripts to full-scale simulations across hundreds of processors—without rewriting your core model.

Debugging and Monitoring Tips#

  • Use print or puts statements to identify which process is active

  • Check Slurm .out/.err files or use squeue to monitor job status when logged into a login node on the HPC

  • Use Tapis job history panel or t.jobs.getJobStatus() to track progress

  • In OpenSeesMP and OpenSeesSP:

    • use getPID and getNP

  • In Python:

    • Get rank with ops.getPID() or MPI.COMM_WORLD.Get_rank()

    • Use os.getcwd() and os.listdir() to inspect files during execution

Contents