Towards a Sustainable World Realized by Kubernetes and Composable...
summary
Claims that Kubernetes plus composable disaggregated infrastructure (CDI) can attach and detach accelerators on demand across a PCI/CXL switch to raise throughput for peak AI inference and power off idle devices to cut energy use. Describes an edge use case processing security camera streams that scales GPUs by Dynamic device scaling between daytime and nighttime, details work on CDI operators and CRD extensions with OSS communities, and shows a first CDI server demo that functions.
AI-written summary · about
related talks
38mDynamic Scaling of GPUs for Container Apps with Composable Disaggregated...
05mKeynote: The Future and Value of Composable Disaggregated Infrastructure
30mCoHDI: A CNCF Sandbox Project for Dynamic Hardware Composability in Kubernetes
05mProject Lightning Talk: CoHDI: Dynamic Hardware Composition for AI‑Era Kubernetes…
41mDynamic GPU Provisioning in Kubernetes With Composable Disaggregated...
07mData Gravity and Kubernetes: Managing Large-Scale Data Ingest With Minimal Latency
03mThe Next Evolution of Kubernetes: GPU-Centric Infrastructure for AI Workloads
24mUsing Kubernetes with Data Processing Units to Offload Infrastructure