Project Title: Raspberry Pi Cluster v2

Project Overview

Raspberry Pi Cluster v2 is a compact, portable rebuild of v1 with a similar software environment but a much smaller physical footprint. The goal was to keep the familiar SLURM + MPI workflow while making the cluster easier to carry and faster to set up. A live status page tracks node availability and health for quick checks while on the go.

• Cluster Status Monitor: Cluster Status

Objectives

• Portability: Shrink the overall size and power requirements while keeping core capabilities from v1.
• Environment Parity: Maintain a similar SLURM + OpenMPI stack to reuse scripts and workflows.
• Visibility: Provide a lightweight monitoring page for quick status checks and diagnostics.

Hardware Components

• Master Node: 1 x Raspberry Pi (control node for scheduling and monitoring)

• Compute Nodes: Raspberry Pi Zero 2 units for low-power, compact compute

• Cluster Interface: Cluster HAT for rapid deployment and reliable multi-node connections

• Thanks to Cluster HAT: Cluster HAT

Software Configuration

• Operating System: Lightweight Linux distribution optimized for low resource usage
• SLURM: Scheduling and job control across compute nodes
• OpenMPI: Parallel execution for MPI jobs
• Monitoring: A minimal web dashboard for node status and quick health checks
• Visualization: ParaView for reviewing and presenting simulation output

Implementation Details

1. Compact Assembly: Designed a smaller enclosure for easy transport and quick setup.
2. Environment Replication: Matched the v1 software stack for compatibility with existing scripts and workflows.
3. Cluster HAT Integration: Leveraged Cluster HAT to simplify wiring and reduce setup time.
4. Status Monitoring: Built and hosted a lightweight cluster status page for node health tracking.

Demonstration Workload: FEM Brake Heat Transfer Analysis

To validate the cluster on a realistic engineering task, I used Pi Cluster v2 to run a finite element heat-transfer analysis of a brake component as a parallel computing workload. The cluster handled the distributed job execution, and the resulting temperature field was visualized in ParaView to review how heat propagates through the brake geometry over time.

This workflow demonstrates that the cluster is not just a portable systems build, but also a practical platform for engineering computation, batch execution, and scientific-result visualization.

Challenges and Resolutions

• Tighter Resource Constraints: Raspberry Pi Zero 2 nodes required careful tuning of MPI job sizes and concurrency.
• Memory and CPU Limits: Adjusted scheduling defaults, job limits, and MPI settings to avoid oversubscription.
• Portability vs. Performance: Balanced size reduction with enough compute power for meaningful parallel workloads.

Outcomes

• Portable Cluster: A compact, carry-ready cluster with a familiar v1 workflow.
• Improved Mobility: Faster setup and teardown without losing core capabilities.
• Operational Visibility: Live monitoring for quick validation and troubleshooting.
• Parallel Simulation Demo: Successfully used the cluster for a brake FEM heat-transfer job and visualized the results in ParaView.
• Portfolio Deployment: Currently responsible for building and deploying my personal portfolio website every time I update it.

Future Plans

• Power Profiling: Measure and optimize power draw for longer off-grid runs.
• Job Templates: Create MPI job presets for the Zero 2 resource profile.
• Thermal Improvements: Add passive cooling options to maintain stable performance.

This project focuses on portability and practical usability while preserving the distributed computing workflow from v1.