Dissecting an HPC data center, hardware, and the ground-level software that drives numerical research

High-Performance Computing (HPC) has become a cornerstone of scientific and industrial advancement, enabling breakthroughs in diverse fields such as drug discovery, material simulations, and neural network research.

However, the intricate web of components that make up an HPC infrastructure often remains obscured, hindering a comprehensive understanding of its inner workings.

This talk aims to dissect the complex tapestry of HPC infrastructure, shedding light on each component and elucidating the intricate relationships that drive computational power to new heights.

The presentation commences with a description of ORFEO, the hardware that enables its usage as an HPC facility, and the challenges that new projects pose to this facility.

Then, the discussion seamlessly moves into the software layer of HPC infrastructure, where it unravels the complexities of orchestrating computational tasks.

Notably, the role of resource managers like SLURM and container orchestration platforms like Kubernetes in efficiently allocating and managing computing resources is explored.

The talk addresses the challenges and advantages of these systems in coordinating parallel computations, optimizing job scheduling, and adapting to dynamic workloads.

Moving beyond resource management, exploring storage solutions and identity management solutions underscores the critical role they play in serving data and securing access within an HPC environment.

Finally, this talk presents VirtualOrfeo, the first iteration of a realistic virtual environment to play with the discussed technologies.

Speaker: Ruggero Lot, RIT – Area Science Park