All projects

workflowsim-research · Research software · Java

Scientific workflow
simulation.

From declared assumptions to inspectable results.

A discrete-event simulation environment for studying how scientific workflows are scheduled. Built on the WorkflowSim and CloudSim codebase, the project connects workflow inputs, resource models, scheduling decisions, and experiment evidence.

The problem

A scheduling result depends on more than the scheduling algorithm. Task dependencies, data availability, resource capacities, and the rules of the simulation all affect what can happen and when.

Studying those decisions requires an explicit experiment setup and a way to trace results back to that setup. This project brings the model, execution, and evidence together so that a comparison can be understood in context.

The approach

Declare the experiment.

Workflow inputs, platform settings, and scheduling choices are provided explicitly. Validation checks inputs and supported combinations before a simulation runs.

Keep computation and data connected.

The model covers task dependencies, compute resources, storage, and data transfer. The supported combinations are documented so each experiment has a defined scope.

Keep the evidence with the result.

Each run has a dedicated output directory. Manifests, metrics, event records, and offline reports help inspect what happened; replay checks compare recorded results within the declared configuration.

Follow an experiment

The experiment workbench connects configuration, execution, and reporting. A run follows this sequence:

  1. Declare

    Choose a workflow, platform, and supported scheduling configuration.

  2. Validate

    Check inputs, parameters, and model compatibility.

  3. Simulate

    Execute scheduling decisions, task events, and data movement.

  4. Inspect

    Read the report and examine the evidence behind the result.

The offline report includes scheme comparisons, task timelines, waiting time, and resource utilization. The repository’s online scheduling example is one starting point for exploring this process.

View the example configuration

Scope matters.

Results describe the declared simulation model and support controlled comparisons under those assumptions. Real infrastructure measurements require their own validation.

The repository provides the simulation environment and policy interfaces, including state, action, and reward contracts. Deep-learning trainers remain external to this project.

Supported combinations and evidence guarantees are documented in the repository. Those contracts provide the context for interpreting a result.

Read the reproducibility contract (Chinese)

Explore the project