Accelerator Docs

Appearance

🧪 Running Scientific Simulations and Experiments ​

The Accelerator platform is designed to streamline the execution and management of scientific simulations and experiments at scale. It provides a unified interface to run, monitor, analyze, and share computational experiments — from standalone jobs to complex jobflows — using powerful Kubernetes-based orchestration.

🎯 Why Use Accelerator for Scientific Workflows? ​

Traditional approaches to scientific computing often face bottlenecks:

  • Manual setup of computing environments
  • Inconsistent code execution across systems
  • Complex dependency handling
  • Lack of reproducibility
  • Difficulty in sharing results with collaborators

Accelerator solves these problems by providing:

✅ Reproducible execution environments
✅ Standardized data exchange between routine
✅ Web and CLI interfaces for orchestration
✅ Built-in data validation and visualization
✅ Collaborative sharing tools
✅ Git + DVC-based code and data versioning

⚙️ Define Your Simulation as a Routine ​

At the core of Accelerator is the concept of a Routine — an encapsulated unit of computation.

To create your simulation routine:

  • Source your code from:

    • Local folder
    • Remote Git repository
    • Container image

  • Define environment requirements via:

  • Specify the routine logic and resource requirements in a wkube.py

🧰 See Getting Started with Routine for an example.

🔁 Combine Simulations Using Jobflows ​

Scientific workflows often consist of multiple dependent steps. Accelerator supports defining acyclic graphs of routines, called Jobflows, allowing:

  • Multi-step simulations
  • Parallel ensemble runs
  • Sequential post-processing
  • Flexible branching and callbacks

You can:

  • Create jobflows via code using add_child() and add_callback()
  • Or use a visual GUI-based drag-and-connect system

📘 Learn more in Jobflow Documentation

🧩 Versioning Code and Data with Git + DVC ​

Reproducibility is critical in scientific research. Accelerator supports Git + DVC-powered versioning:

  • Use the built-in routines:
    • 🟢 DVC-powered Git Push — Store datasets and code snapshots
    • 🔵 DVC-powered Git Pull — Retrieve previous versions for re-use

These routines ensure your experiments are:

  • Fully version-controlled
  • Easily reproducible
  • Transparent for collaboration and review

🔄 Refer to Inbuilt Routines Guide for DVC integration examples

📁 Share Data Across Simulations ​

Simulations frequently need to exchange data between steps. Accelerator enables this through its **data mapping system **:

  • Inputs and outputs can be passed via:
    • Cloud-based paths (acc://)
    • Mounted shared volume (/mnt/pipe)
    • Selected files/folders from GUI

📦 See Data Mapping Guide

📊 Validate and Visualize Scientific Data ​

Scientific simulations often generate structured data that must be validated for consistency and quality.

Accelerator provides:

  • Built-in data validation schemas
  • Custom rule-based validation templates
  • Automatic visualization for supported data types

You can validate experimental outputs and view results directly in the browser using charts or map-based explorers.

🧪 Explore Data Validation Guide
📊 See Data Visualization Tools

🌐 Collaborate and Reuse ​

Each simulation or routine can be:

  • Saved as a reusable routine entity
  • Shared with collaborators within the project space
  • Triggered manually or automatically
  • Embedded as interactive visual dashboards

This makes Accelerator ideal for team-based research, reproducible experiments, and knowledge transfer.

💡 Refer to GUI-Based Routine Interface for entity creation

🧬 Use Cases in Scientific Contexts ​

DomainExample Applications
Climate ModelingDownscaling, scenario analysis, ensemble simulations
Earth ObservationRaster data processing, reclassification, harmonization
AgricultureYield estimation models, data fusion pipelines
Energy SystemsRegional projections, technology portfolio simulations
Health & PolicyExposure modeling, epidemiological simulations

🔐 Build Trust Through Provenance ​

By combining versioned code, structured metadata, and repeatable data flows:

  • You gain auditability of every experiment
  • Collaborators can verify results
  • Stakeholders can trust insights

Accelerator makes it easy to track what was run, with what data, on what codebase, and when.

🚀 Summary ​

Accelerator transforms simulation workflows by:

  • Encapsulating them as modular routines
  • Validating inputs and standardizing outputs
  • Scaling execution on Kubernetes seamlessly
  • Versioning both code and data with Git + DVC
  • Enabling interactive visual review of results
  • Facilitating reuse, collaboration, and reproducibility

Whether you're running experimental code or publishing a stable model, Accelerator provides the infrastructure to * scale science reliably*.