“This approach is commonly implemented using Kubernetes, GPU device plugins, and workload schedulers.”
+ resource requests and limits support
“Container platforms such as Kubernetes support resource requests and limits.”
+ GPU resource allocation via device plugins
“Kubernetes supports GPU resource allocation through vendor-specific device plugins.”
+ quota enforcement
“Kubernetes tracks resource consumption and rejects new workloads when quotas would be exceeded.”
− no automatic fractional GPU memory limits
“It doesn't automatically enforce arbitrary fractional GPU memory limits.”
“- User A gets an NVIDIA A100.”
“- Prometheus: Stores and collects metrics.”
“- Grafana: Displays dashboards.”
“Notebook environments | JupyterHub”
“AWS Documentation”
