What is Fuzzball? Container Orchestration Built for HPC and AI Workloads

This short explainer from David Godlove sets out the problem Fuzzball exists to solve. Teams now need to run traditional HPC applications such as molecular dynamics, fluid dynamics and weather modeling on the same infrastructure as machine learning training and interactive Jupyter notebooks, and traditional HPC schedulers were not built for that mix.

Container orchestration is the missing piece, but Kubernetes brings overhead and forces researchers to learn microservice concepts they do not need. Fuzzball is container orchestration built specifically for performance-intensive computing, handling batch HPC workloads alongside AI training and interactive jobs with the simplicity researchers expect and the controls administrators require.

Key takeaways

• HPC and AI workloads are converging, with molecular dynamics, fluid dynamics and weather modeling sharing infrastructure with ML training.
• Traditional HPC schedulers were not designed to handle this mix of batch, AI training and interactive workloads.
• Kubernetes adds overhead and requires researchers to learn microservice concepts that are irrelevant to their work.
• Fuzzball is container orchestration purpose-built for performance-intensive computing, covering HPC, AI training and interactive jobs.

Questions this video answers

What is Fuzzball?

Fuzzball is container orchestration built specifically for performance-intensive computing. It runs traditional HPC workloads and modern workflows such as AI training and interactive jobs like Jupyter notebooks on one platform, offering the simplicity researchers expect and the controls administrators require, without the overhead and microservice concepts that come with Kubernetes.

About this video

HPC and AI workloads are converging — and your infrastructure needs to keep up. Researchers need to run molecular dynamics, fluid dynamics, and weather modeling alongside ML training jobs and interactive Jupyter notebooks, all on the same cluster. Traditional HPC schedulers weren't built for this. And Kubernetes adds microservice complexity that researchers shouldn't have to deal with. In this video, CIQ's David Godlove explains what Fuzzball is and why it exists:

The growing gap between what traditional HPC schedulers can handle and what modern research teams actually need Why Kubernetes isn't the right answer for performance-intensive computing environments How Fuzzball delivers container orchestration purpose-built for HPC — handling traditional batch jobs, AI training workloads, and interactive sessions like Jupyter notebooks on a single platform How Fuzzball gives researchers the simplicity they expect while giving administrators the policy controls they require

If you're evaluating how to modernize your HPC infrastructure without forcing your researchers to become Kubernetes experts, this is the place to start.

This video is part of the Fuzzball playlist. Browse every CIQ video by product and topic.