Workflow Design

Workflow Design#

Designing Computational Workflows Around Your Research Needs

Your computational workflow should be designed around your research needs, not around a specific tool, platform, or computing system. Different research questions impose very different computational requirements, and those requirements naturally lead to different workflow designs. A workflow that works well for exploratory analysis may fail completely when scaled to thousands of simulations—or when extended to include coupled physics, uncertainty quantification, or machine learning.

The first step in designing an effective workflow is therefore to identify what you are trying to compute, not where you intend to run it.

A Practical Way to Think About Workflow Design#

Rather than building a single, rigid pipeline, it is far more effective to break your workflow into reusable modules, each responsible for a well-defined task. Common modules include:

  • Input generation (models, parameters, ground motions, meshes)

  • Execution (single simulations, ensembles, training loops)

  • Post-processing (response extraction, aggregation, visualization)

  • Iteration or automation (parameter sweeps, Monte Carlo loops, retraining)

Once your workflow is modular, you can:

  • Reuse the same components across different projects

  • Swap execution environments as your computational demands grow

  • Combine modules in new ways as your research questions evolve

This modular approach is what allows workflows to remain adaptable over time, instead of being rewritten from scratch whenever requirements change.

Why Scalability Must Be Designed In From the Start#

Scalability is not something you “add later.” It must be considered during workflow design.

A workflow that runs successfully for:

  • one model

  • one parameter set

  • one dataset

should be able to scale to:

  • hundreds or thousands of simulations

  • larger or higher-resolution models

  • more complex coupling or uncertainty

However, scaling is not just about adding more nodes. Different analysis types scale in different ways:

  • Some workloads scale trivially by running many independent jobs.

  • Others are limited by memory per node, communication costs, or solver structure.

  • Some benefit from GPUs; others do not.

DesignSafe supports this diversity by providing multiple execution environments, but it is your workflow design that determines whether those resources can be used effectively.

Key takeaway: On DesignSafe, HPC resources are not just about scaling “up” by adding more nodes — they are also about adapting to the computational profile of your analysis. Each workload has its own “best fit” execution environment.

Contents