07. Cloud & Scheduling

Chapter focus: This chapter explains how Slurm and Kubernetes can work together in AI-HPC production environments.

Focus Areas

• workload scheduling and isolation
• containerized training workflows
• multi-cluster collaboration and elasticity

🧭 Reading Map

• Prerequisites: 05. Provisioning — a version-consistent base first, then scheduling
• Role: Part 2 scheduling chapter — Slurm and Kubernetes: traditional HPC jobs vs. cloud-native inference
• Order: Slurm line (install→practice→user guide) → K8s line (GPU scheduling→vLLM production)
• Next: 08. Heterogeneous Computing — schedulers serve more than NVIDIA

Articles

Kubernetes GPU Scheduling: From Device Plugin to Shared GPUs

Reproducible manifests dissecting K8s GPU scheduling: how Device Plugin works, shared-GPU options (time-slice/MIG/MPS), GPU monitoring and quotas, and the K8s-Slurm division of labor, with engineering takeaways.

阅读文章 →

Rocky Linux 9 HPC Cluster Deployment

A full-stack guide to deploying an HPC cluster on Rocky Linux 9.4, covering NFS, MUNGE, MariaDB, Slurm 24.05, and Prometheus monitoring.

阅读文章 →

Slurm Practice: Distributed Training with Docker

A comprehensive guide on submitting distributed training jobs with Docker containers via Slurm, featuring a complete Megatron-LM job script.

阅读文章 →

Ubuntu 22.04 + Slurm 24.05 Cluster Deployment

A practical guide to deploying a GPU cluster on Ubuntu 22.04, covering Netplan, NVIDIA Drivers, Slurm 24.05 compilation, Cgroup v2, and PyTorch testing.

阅读文章 →

Slurm User Guide: Job Submission & Management

A comprehensive manual for Slurm users, covering basic job scripts, multi-node distributed training templates, and essential commands.

阅读文章 →

Kubernetes Inference Service in Production: vLLM from Deployment to High Availability

orchestration side: GPU quotas, rolling updates, graceful shutdown, HPA, monitoring.

阅读文章 →

K8s AI Scheduling: GPU Sharing, Topology & Elasticity

Three abstraction layers, topology-aware scheduling & comm amplification, inference elasticity vs training queues, Slurm positioning.

阅读文章 →

Kubernetes + RAGFlow Deployment Guide

A comprehensive guide to deploying a single-node Kubernetes cluster and RAGFlow application on Ubuntu 22.04 VM.

阅读文章 →

NVIDIA PyTorch Docker Deployment Guide

A complete guide to pulling, configuring, and running NVIDIA PyTorch containers, including SSH setup, volume mounting, and VSCode remote debugging.

阅读文章 →