Serverless Kubernetes Boosts AI Business
summary
Kubernetes-backed Cloud Container Instance (CCI) is a serverless layer that runs AI workloads with kata containers for tenant isolation, supports NVIDIA GPUs and InfiniBand networking for distributed training, and reports being 95% API compatible with upstream Kubernetes. The platform extends the scheduler to do fine-grained GPU-aware placement, exposes GPU drivers from hosts into containers, handles model image sizes (tens of GB) via fast storage and object stores, and experiments with GPU sharing and NVidia MIG-style partitioning.
AI-written summary · about
chapters (5)
- 0:00Start
- 0:37Serverless Computing?
- 9:34Advantages
- 29:21Scheduler - Poseidon
- 30:45Scheduler - Benchmark of Poseidon
related talks
05mContainerization Beyond CPUs – A Kubernetes-based Serverless Pl...
27mHarnessing Kubernetes for Scalable AI/ML Workloads
16mAI + Kubernetes: What Beginners...
30mRunning AI Workloads in Containers and Kubernetes
21mAIaaS with Dell AI ready infrastructure
23mThe State of Cloud Native: The Shift Towards AI
03mThe Next Evolution of Kubernetes: GPU-Centric Infrastructure for AI Workloads
38mKubernetes for AI Workloads – What Works and What… Doesn’t
