Skip to main content

Introduction

Drona Workflow Engine is an HPC workflow assistant and framework developed by Texas A&M University HPRC. It provides a 100% graphical interface for creating and submitting computational jobs — researchers provide workflow-specific information through dynamic forms, and Drona generates all necessary scripts, validates the configuration, and submits the job on their behalf.

The Problem​

HPC resources are becoming increasingly complex while simultaneously growing more popular among researchers who may lack traditional HPC skills. Writing Slurm scripts, managing modules, configuring multi-stage pipelines — these tasks create a steep learning curve that leads to frustration, inefficient resource usage, and increased burden on HPC support teams.

How Drona Solves It​

Drona abstracts HPC complexity behind intuitive, workflow-specific forms while keeping the researcher in full control.

Guided Workflow Configuration​

Researchers select a workflow environment and fill out a dynamic form tailored to that specific use case. The form adapts based on user selections — fields appear, hide, and validate in real time. No Slurm syntax or HPC knowledge required.

Drona Input Form

Full Transparency​

Before submission, Drona displays all generated scripts in a fully editable preview window. Researchers can review exactly what will run, make adjustments, and catch issues before they reach the scheduler. A message pane provides validation warnings and environment-specific guidance.

Drona Preview Window

Extensible Framework​

Drona is not just an assistant — it's a framework. Workflows are defined declaratively through a small set of configuration files:

  • schema.json — Defines the form interface and field behavior
  • map.json — Maps form values to job script variables
  • driver.sh — Template for the generated job script
  • utils.py — Optional custom processing logic

Anyone can create, customize, and share workflows. System administrators deploy workflows for all users, while researchers can develop and test their own in a personal workspace.

Dynamic Content​

Workflows can include retriever scripts that fetch live data — available GPU types, module versions, running job statuses — and populate form fields dynamically. This enables workflows that respond to the current state of the cluster.

Validation and Feedback​

Drona analyzes input values and provides warnings before submission. For example, it might flag that a GPU was requested for longer than the cluster's maximum GPU wall time, or that a required module is not available on the selected cluster.

Who Is Drona For?​

  • Researchers who want to run scientific workflows on HPC without learning Slurm
  • HPC administrators who want to provide guided, validated job submission for their users
  • Workflow developers who want to build shareable, reusable computational environments

Next Steps​

Texas A&M University High Performance Research Computing