Skip to content

Deployment

Quantum Pipeline supports multiple deployment strategies, ranging from a lightweight PyPI installation for development to a full Docker Compose deployment that includes the entire data platform. This section covers each approach in detail, along with GPU acceleration configuration and environment variable reference.

Deployment Options

Deployment Services Use Case
PyPI Package Quantum Pipeline only Development, quick experiments
Docker (single container) Quantum Pipeline (CPU or GPU) Isolated simulations
Docker Compose Full platform (all services) Production, thesis experiments

The PyPI package is the simplest option. Install with pip install quantum-pipeline and run simulations directly. This does however require the user to set up other services as well as ensure connectivity.

A single Docker container provides an isolated environment with all Python dependencies pre-installed. The GPU variant includes CUDA libraries and a custom-built Qiskit Aer with GPU support. This is suitable for running simulations without the full data pipeline.

The Docker Compose deployment brings up the complete platform: Quantum Pipeline containers (CPU and GPU), Apache Kafka with Schema Registry, Spark cluster, Airflow orchestrator, MinIO object storage, and monitoring services. This is the configuration used for thesis experiments.

Resource Allocation

The thesis experiments used a system with an Intel Core i5-8500 (6 cores), 56 GB RAM, and two NVIDIA GPUs (GTX 1060 6GB + GTX 1050 Ti 4GB).

It was stable and functioned well for extended periods of time. I'd imagine minimal requirements are much lower - naturally, while accepting longer times of iteration.

Guides

  • Docker Basics


    Container images, building from source, and running individual containers.

    Docker Basics

  • Docker Compose Deployment


    Full platform deployment with all services, networking, and health checks.

    Docker Compose

  • GPU Acceleration


    NVIDIA driver setup, CUDA configuration, and performance benchmarks.

    GPU Acceleration

  • Environment Variables


    Complete reference for all configuration variables used across services.

    Environment Variables