nebulnetes

Nebulnetes (n8s)

Nebulnetes is the open-source platform for running AI workloads on Kubernetes in an optimized way, both in terms of hardware utilization and workload performance.

Product design

Nebulnetes design relies on 4 layers, each focusing on specific aspects of the AI/ML ecosystem and infrastructure, with the aim of providing a full stack for running and managing AI workloads in the most efficient and easy way possible.

1. AI Application [🚧WIP]: layer that integrates libraries and framework from the AI and ML ecosystems and from which the different kind of AI workloads are generated (notebooks, batch training jobs, ML pipelines, etc.).
2. Gateway [🚧WIP]: layer focused on the governance and monitoring of the AI workloads generated by the application layer
3. Operating system: layer that schedules workloads and hides the complexity of the underlying hardware
4. Hardware: layer that provides the necessary resources and accelerators to run workloads in the optimal way

Operating system

The operating system layer is responsible for workloads scheduling and hardware abstraction. It orchestrates the workloads taking into account considerations specific for AI/ML workloads and leveraging techniques typical of High-performance Computing (HPC), and it hides the underlying hardware complexities.

Currently, this layer provides two features Automatic GPU partitioning and Elastic Resource Quota management.

Documentation

• Automatic GPU partitioning
  • Getting started
  • Enable nodes for automatic partitioning
  • MIG Partitioning
  • Configuration
  • Integration with Nebulnetes scheduler
  • Troubleshooting
• Elastic Resource Quota management
  • Getting started
  • How to define Resource Quotas
  • Installation options
  • Troubleshooting

Developer

• Getting started
• Contribution guidelines
• Roadmap

Join the community | Contribute