Kubeflow: Multi-Tenant, Self-Serve, Accelerated Platform for Practitioners
Talkat KubeCon + CloudNativeCon North America 2019 · on Wednesday, 2019-11-20· at 15:20· Room 17AB - San Diego Convention Center, Mezzanine Level·
summary
Kubeflow adds a profile CR that creates a per-user workspace namespace with owner/invite/revoke controls, a UI and CLI for self-service, and admin-resettable policies and quotas; identity plugins map users to cloud service accounts and admission/controllers inject acceleration overlay specs (CPU family, GPU count) so jobs land on the right nodes. Demo shows notebook spawn, invite/revoke syncing in ~10–30s, pipelines and Intel DAAL/MKL‑DNN (TensorFlow 1.14+) integrations; all components are open source.
AI-written summary · about
chapters (9)
- 0:00Start
- 0:28Enabling the end user
- 2:05Kubeflow: A platform for building ML products
- 4:57Multi Tenancy for End Users
- 6:26Kes Multi-Tenancy Challenges
- 8:14Kubeflow Multi-Tenancy (Profiles)
- 10:59Kubeflow Access Control
- 14:56Share Kubeflow Profile with other users
- 19:56Tekton Pipeline to run a model
related talks
14mKubeflow: Where ML Meets Kubernetes
35mKubeflow: Data Scientist ready?
31mKubeflow - Data Science on Steroids
21mIntroducing Kubeflow (w. Special Guests Tensorflow and Apache Spark)
27mKubeflow for enabling AI Powered Drug Discovery and development...
29mKubeflow: TensorFlow on Kubernetes
30mKubeflow, Your ML Tools on Kubernetes
31mKubeflow Intro – Michał Jastrzębski & Ala Raddaoui, Intel