Introduction to Container Usage in HPC

• Understand the role of containers in HPC environments
• Distinguish between Docker and HPC-oriented container solutions
• Run and manage containerized workloads on HPC systems
• Integrate containers with GPU resources and job schedulers
• Apply debugging and lifecycle management techniques

Session 1 – HPC Architecture & Containers (45 min)

• Fundamentals of HPC architecture
• Why containers in HPC?
• Docker vs HPC-oriented solutions
• Security considerations in shared environments

Session 2 – Working with Apptainer (45 min)

• Container image formats
• Pulling and building container images
• Running containers in HPC
• Bind mounts and filesystem integration

Session 3 – GPU & Scheduler Integration (45 min)

• GPU passthrough in containers
• Multi-GPU execution
• Slurm integration
• Running batch jobs with containers

Session 4 – Customization & Debugging (45 min)

• Installing packages inside containers
• Writable overlays and persistent storage
• Common runtime errors and debugging strategies
• Best practices for reproducible HPC workflows

Dr. İsmail Güzel, Senior Researcher, TRUBA HPC Centre, TÜBİTAK ULAKBİM

Dr. İsmail Güzel is a Senior Researcher at the TRUBA HPC Centre of TÜBİTAK ULAKBİM, specializing in GPU-accelerated artificial intelligence and high-performance computing. He holds a Ph.D. in Mathematics Engineering from Istanbul Technical University, with research focused on Topological Data Analysis and Machine Learning. At TRUBA, he provides expert-level support for large-scale, multi-node, multi-GPU AI workloads, including LLM training and computer vision systems. He teaches GPU computing within the National Technology Initiative – Artificial Intelligence Specialization Programme and contributes as a researcher to the BSC AI Factory project. He holds multiple NVIDIA certifications in deep learning and GPU programming.

Container technologies have become essential for reproducible and portable scientific computing workflows. However, traditional container tools such as Docker are not directly suitable for multi-user HPC systems due to security and architecture constraints.

This course introduces participants to HPC-compatible container solutions such as Apptainer, explaining how they differ from Docker and how they integrate with GPUs and job schedulers. Through practical demonstrations on the TRUBA HPC infrastructure, participants will gain hands-on insight into running AI and scientific workloads inside containers in a production-grade HPC environment.

The course combines conceptual foundations with applied demonstrations, enabling participants to independently deploy and manage containerized workflows on HPC systems.

Prior experience with HPC systems and/or scientific computing environments.