Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/dominodatalab/nvidia-starter-kit
Utilities for installing Domino on a single NVIDIA GPU node
https://github.com/dominodatalab/nvidia-starter-kit
Last synced: 29 days ago
JSON representation
Utilities for installing Domino on a single NVIDIA GPU node
- Host: GitHub
- URL: https://github.com/dominodatalab/nvidia-starter-kit
- Owner: dominodatalab
- Created: 2020-11-24T20:16:19.000Z (about 4 years ago)
- Default Branch: main
- Last Pushed: 2021-01-12T20:04:10.000Z (almost 4 years ago)
- Last Synced: 2023-08-07T03:05:38.967Z (over 1 year ago)
- Language: Shell
- Size: 65.4 KB
- Stars: 0
- Watchers: 5
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
## Domino NVIDIA Starter Kit
Domino NVIDIA Starter Kit provides a fast and simple installation of Kubernetes, [Domino](https://www.dominodatalab.com/), and all other dependencies on a single node for
use in demos of Domino software on [DGX hardware](https://www.nvidia.com/en-us/data-center/dgx-systems/) with [Multi-Instance GPU A100 modules](https://www.nvidia.com/en-us/technologies/multi-instance-gpu/).
For this project, we're using the [NVIDIA DeepOps](https://github.com/NVIDIA/deepops) tooling to install hardware
drivers, the NVIDIA container runtime, and Kubernetes 1.18.
Domino 4.3.3+ can then be installed from official release channels via the [fleetcommand-agent](https://admin.dominodatalab.com/en/4.3.2/installation/install.html).
### How to use Domino NVIDIA Starter Kit
- Quickly stand up Domino as an ML development and deployment layer on a single DGX machine
- Demo Multi-Instance GPU capabilities on a machine with A100 Tensor Core modules, such as a p4.24xlarge EC2 instance
### How NOT to use Domino NVIDIA Starter Kit
- Since Domino NVIDIA Starter Kit deployments have no redundancy or backups, they should not be treated as durable production data science platforms
- Domino NVIDIA Starter Kit can not currently be used to build multi-node clusters
### Contents
* [Installation prerequisites](#installation-prerequisites)
* [Kubernetes and dependency installation](#kubernetes-and-dependency-installation)
* [Domino installation](#domino-installation)
### Installation prerequisites
- **Infrastructure**
This process is only suitable for installing Domino on a **single NVIDIA GPU node**. The node should be running a
DeepOps-compatible OS (Ubuntu 18.04 recommended) with Git installed, have [unencumbered access to Domino release
channels over the Internet](https://admin.dominodatalab.com/en/4.3.2/installation/install.html#requirements), and have eight
[NVIDIA A100 Tensor Core GPU](https://www.nvidia.com/en-us/data-center/a100/) modules installed (other architectures coming soon).
Record the IP address or hostname of this machine. You'll need it when configuring DNS for the Domino application.
- **DNS**
Domino is a web application with multiple interlinked UI components, and must accordingly have a fully qualified domain
name with which to self-reference. The installation process will generate a self-signed certificate for this name and
configure Domino to serve it. Ideally, this would be a name for which you can control DNS for your organization,
however you can use any name and override DNS in your `/etc/hosts` file if preferred.
Record the FQDN you want to use to access Domino. You'll need it when running the Domino software installer.
- **Credentials**
In order to download Domino component images from [Quay.io](https://quay.io/) release channels, you will need a username and password
provided by Domino. If you do not know your username and password, contact Domino at [email protected].
Record these credentials. You'll need them when running the Domino software installer.
### Kubernetes and dependency installation
This command installs Kubernetes 1.18 plus NVIDIA GPU drivers and utilities on the node. The machine will restart twice during this installation.
`git clone https://github.com/dominodatalab/nvidia-starter-kit.git && cd nvidia-starter-kit/ && ./domino-nvidia-starter-kit-1.sh`
Once this completes, you are ready to install Domino.
### Domino installation
Run this script with the arguments shown from the `nvidia-starter-kit` directory to install Domino 4.3.3 into the local node Kubernetes environment.
`./install-domino.sh `
Once the installation completes, you can access Domino by creating DNS records (or /etc/hosts entries) that
resolve the Domino FQDN to the IP address of your node. To get started, browse to a path of `/auth/`
(the trailing `/` is important) on your Domino FQDN to open the authentication service. Then follow
[these instructions in the Domino administrator's guide](https://admin.dominodatalab.com/en/4.3.2/keycloak.html) to
retrieve credentials and log in to the auth service, from which you can create users or enable independent signups.