Empowering Data Science with Kubernetes | Samuele Chiocca and Andrea Rota
summary
Tenaris migrated its data-science platform onto Kubernetes (Fury by SIGHUP) with a cluster of three masters, about 10 workers, and two bare-metal NVIDIA GPU nodes. They provide per-user persistent JupyterHub/JupyterLab notebooks on NFS, use Drone + Nexus plus kustomize for CI/CD, evaluated but skipped Kubeflow, and achieved faster model rollout (days to 1–2 hours), quicker updates (minutes), reproducible isolated environments, GPU visibility via nvidia-smi, and an on-prem platform for interactive analysis and serving.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 3:49Data Science at Tenaris
- 15:26Kubernetes architecture
- 23:08Kubernetes tools demo
- 30:54Benefits and conclusions
- 33:46Q&A
related talks
39mHow We Hosted 200+ Data Scientists in a K8S Cluster
27mHarnessing Kubernetes for Scalable AI/ML Workloads
31mKubeflow - Data Science on Steroids
37mA Day in the Life of a Data Scientist. Conquer ML Lifecycle on Kubernetes
54mEconomics and Best Practices of Running AI/ML Workloads on Kubernetes
46mMachine Learning with Kubernetes
34mThe Data Analytics Platform or How to Make Data Science in a Box Possible
40mStreet fighting techniques for multi tenant machine learning...