KAI Scheduler Demos

This repository contains demonstration examples for the KAI Scheduler, showcasing its capabilities in both CPU and GPU resource management. These demos illustrate how the scheduler handles complex scheduling scenarios, including priority-based scheduling, workload consolidation, and gang scheduling.

Table of Contents

• Getting Started
• Demo Contents
  • GPU Scheduling Example
• Prerequisites
• Documentation

Getting Started

Each demo has its own README with detailed instructions. Please refer to:

• GPU Scheduling Demo Guide

Demo Contents

Scheduling Example for GPU

• Demonstrates GPU resource allocation and management
• Shows priority-based scheduling
• Includes a distributed training workload
• Demonstrates gang scheduling capabilities
• Shows resource consolidation for GPU workloads

Prerequisites

To run these demos, you need:

1. A Kubernetes cluster with KAI Scheduler installed (version 1.0 or later)
2. 4 nodes with 4 GPUs each (16 total GPUs)
3. NVIDIA device plugin installed and configured
4. Kubeflow Training operator installed

Documentation

For more detailed information about features and how the KAI scheduler works, please refer to the documentation. We welcome any questions in Github issues!