Flock HPC Lab

A Docker-based lab environment to practice and familiarise oneself with the principles of High Performance Computing (HPC), cluster administration, workload management, and Linux infrastructure.

---

Overview

Flock is a small-scale HPC cluster built as a reproducible laboratory environment.

The goal is to learn and practise the core concepts behind HPC systems:

• Linux-based compute environments
• Cluster architecture
• Job scheduling
• Resource allocation
• Batch processing
• Interactive workloads
• Authentication between nodes
• Shared storage concepts
• Accounting and monitoring
• Containerised infrastructure

Although the physical resources are modest, the architecture follows the same principles used by much larger research computing environments.

---

Architecture

                    User
                     |
              SSH login node
              flock-login
                     |
              Slurm controller
              slurmctld
                     |
          +----------+----------+
          |                     |
     compute001            compute002
       slurmd                slurmd
          |                     |
          +----------+----------+
                     |
              Job execution
                     |
              MariaDB accounting

---

Components

Component Purpose

---

flock-login User access and administration slurmctld Slurm workload scheduler compute001 Compute node compute002 Compute node MUNGE Cluster authentication MariaDB Persistent accounting database Docker Compose Orchestration Colima Local virtualisation

---

Technology Stack

• Ubuntu 24.04 LTS
• ARM64
• Docker
• Docker Compose
• Colima
• Slurm 23.11
• MUNGE
• MariaDB 11

---

Why Flock?

Modern HPC environments combine many services:

• login nodes
• scheduler nodes
• compute nodes
• authentication systems
• storage
• databases
• monitoring

Flock provides a local environment to explore these concepts without requiring access to a production cluster.

---

Slurm Examples

Submit a job:

sbatch job.slurm

View jobs:

squeue

Inspect jobs:

scontrol show job <JOBID>

View cluster status:

sinfo

Interactive allocation:

salloc

---

Example Batch Job

#!/bin/bash

#SBATCH --job-name=test
#SBATCH --partition=long
#SBATCH --time=01:00:00
#SBATCH --mem=512M

hostname
sleep 300

Submit:

sbatch test.slurm

---

Job Lifecycle

User
 |
sbatch
 |
slurmctld
 |
Scheduler
 |
Compute node
 |
slurmd
 |
Job execution
 |
Accounting

---

Authentication

Flock uses MUNGE authentication between cluster components.

This provides:

• secure credentials
• node authentication
• identity verification

---

Storage Model

Directory Purpose

---

/home User configuration and files /project Shared project data /scratch Temporary working area

---

Accounting

Flock includes MariaDB as the foundation for Slurm accounting.

slurmctld
    |
 slurmdbd
    |
 MariaDB

This allows historical analysis of:

• completed jobs
• CPU usage
• memory usage
• workloads

---

Project Structure

flock/
├── compose.yml
├── config/
│   ├── slurm.conf
│   └── cgroup.conf
├── image/
│   ├── Dockerfile
│   └── entrypoint.sh
├── jobs/
├── secrets/
├── shared/
└── state/

---

Skills Demonstrated

Building Flock provided practical experience with:

• Linux administration
• Docker infrastructure
• virtualisation
• networking troubleshooting
• Slurm administration
• workload management
• database integration
• infrastructure as code
• open-source documentation

---

Future Development

Possible future work:

• full Slurm accounting integration
• Prometheus monitoring
• Grafana dashboards
• job arrays
• workflow dependencies
• GPU simulation
• additional compute nodes

---

Disclaimer

Flock is a learning and development environment designed to demonstrate HPC concepts and cluster administration principles.

It is not intended to replace production HPC infrastructure.

---

Author

Built as a personal HPC learning laboratory exploring Linux systems, scientific computing infrastructure, workload management, and open-source technologies.