Skip to content

Getting Started

This guide will help you set up and run the HPC AI Benchmarking Orchestrator on the MeluXina supercomputer.

Overview

The HPC AI Benchmarking Orchestrator is a Python framework for deploying, benchmarking, and monitoring containerized AI services on HPC clusters. It automates the complex workflow of SLURM job submission, container management, and metrics collection.

Key Capabilities:

  • Deploy AI services (Ollama, Redis, Chroma, MySQL) via SLURM
  • Run automated benchmark workloads with configurable parameters
  • Collect real-time metrics via Prometheus and cAdvisor
  • Visualize performance through Grafana dashboards
  • Generate benchmark reports and analysis

System Requirements

Local Machine

  • Python 3.9+
  • SSH client with key-based authentication
  • Git

HPC Cluster (MeluXina)

  • SLURM workload manager
  • Apptainer/Singularity for containerization
  • Access to GPU nodes (for Ollama)
  • Project allocation (account: p200981 or your project)

Quick Overview

sequenceDiagram
    participant User
    participant CLI as main.py
    participant SSH as SSHClient
    participant SLURM
    participant Container as Apptainer

    User->>CLI: python main.py --recipe service.yaml
    CLI->>SSH: Connect to MeluXina
    SSH->>SLURM: sbatch job_script.sh
    SLURM->>Container: Launch on compute node
    Container-->>SLURM: Service running
    SLURM-->>SSH: Job ID + Node
    SSH-->>CLI: Service info
    CLI-->>User: Service started: abc123

Workflow

  1. Configure - Set up config.yaml with HPC credentials
  2. Deploy - Start services using YAML recipes
  3. Benchmark - Run client workloads against services
  4. Monitor - View metrics in Grafana dashboards
  5. Analyze - Download and process results

Next Steps

Continue to Installation →