HPCv3 is a lightweight simulator for studying job scheduling and resource allocation in High-Performance Computing (HPC) systems.
It allows you to generate workloads, define platforms, convert swf formats to JSON, run scheduling simulations with different algorithms, and visualize results.

• Python 3.9 or newer
• Recommended: create a virtual environment to isolate dependencies

git clone https://github.com/algoritmakomputasi-ugm/HPCv3-AIBT-2025.git
cd HPCv3-AIBT-2025
python setup.py

Workload Generation → Define synthetic job traces.

Platform Generation → Define HPC platform topology and machine states.

Scheduling Simulation → Run the simulation with different schedulers.

Results Visualization → Analyze and visualize scheduling outcomes.

Each stage is provided as a Jupyter Notebook for ease of experimentation.

Notebook: workloads_generator.ipynb

This stage generates job traces that describe:

• Number of jobs
• Arrival times
• Job duration
• Resource requirements

You can configure parameters (e.g., inter-arrival rate, job length distribution) to model different HPC workloads. The output is stored as a .json or .csv file, which will be used as input for the simulation.

Notebook: platforms_generator.ipynb

This stage defines the HPC platform, including:

• Number of compute nodes
• Available DVFS (Dynamic Voltage and Frequency Scaling) profiles
• Power consumption and compute speed
• The generated platform description file is also saved as .json, providing the environment where jobs will be scheduled.

Notebook: SPARS_runner.ipynb

Here you run the simulation by combining:

• A workload file (jobs to run)
• A platform file (system configuration)
• A scheduler of your choice

Available schedulers:

1. Easy Backfilling – Improves utilization by allowing smaller jobs to jump ahead if they don’t delay larger jobs.
2. FCFS (First-Come, First-Served) – Jobs are executed in the order they arrive.
3. Smart FCFS – An enhanced FCFS variant with an early switch-on policy.

The simulation produces CSV logs containing job start/finish times, node allocations, and system events.

Notebook: create_ganttchart.ipynb

This stage transforms raw CSV logs into visual insights. Outputs include:

• Gantt chart → shows job execution timelines and node allocations
• Job statistics → execution time, waiting time, utilization
• Energy consumption analysis (if DVFS/platform states are enabled)

These visualizations help you evaluate the effectiveness of different scheduling policies and platform configurations.

1. Generate a workload → workloads_generator.ipynb
2. Generate a platform → platforms_generator.ipynb
3. Run simulation → HPCv3_runner.ipynb
4. Visualize results → create_ganttchart.ipynb

The notebook swf_to_json.ipynb lets you convert SWF format files into JSON files.

MIT License – feel free to use and extend for research and teaching.
Please provide appropriate credit when using or modifying this project in your own work.