Skip to content
This repository has been archived by the owner on May 11, 2024. It is now read-only.

Latest commit

 

History

History

installation-and-configuration

README.md

Nauta Installation, Configuration, and Administration Guide

This Nauta Installation, Configuration, and Administration guide provides step-by-step instructions for installing, configuring, upgrading, and removing Nauta. This guide also provides an overview of Nauta requirements, configuration options, and administration tasks.

This section discusses the following main topics:

Note: For instructions on configuring the Nauta client, refer to the Nauta User Guide.

Nauta is a software suite that provides a multi-user, distributed computing environment for running deep learning model training experiments. Results of experiments can be viewed and monitored using a Command Line Interface (CLI), Web UI and/or TensorBoard. You can use existing data sets, your own data, or downloaded data from online sources, as well as create public or private folders to make collaboration among teams easier.

Nauta runs on Kubernetes and Docker for scalability and ease of management. Nauta uses customizable templates to remove the complexities of creating and running single and multi-node deep learning training experiments without all the system's complexity and scripting needed with standard container environments.

Nauta Hardware Requirement Overview

Nauta is intended to run on a multi-server Kubernetes cluster. To run Nauta, you will need at least one Master node, and one or more Worker nodes. Nauta is a platform for performing Deep Learning training, and requires robust hardware specifications to run with optimal performance.

Minimal Memory and Disk Requirements

Node Count CPU Memory Disk
Master 1 12 32Gi 200Gi
Worker 2 4 8Gi 200Gi

Note: These requirements are based on our development environments and nctl config command specifications. This configuration has been tested and does function. However, it does indicate that Nauta might work on smaller configurations.

Example of Production Environment Setup

Node Count CPU Memory Disk
Master 1 2x40 384Gi 144TB
Worker 4 2x40 192Gi 1TB

Note: This is just an example of production environment, which is able to run multiple resource demanding experiments at once.

Nauta Installation Procedures

To install Nauta in a 'bare metal' (for example, non-cloud) server environment, you will need to:

  1. Execute the following commands from the command line:

  • git clone --recursive https://github.com/IntelAI/nauta.git

  • cd nauta

  1. Build the base package (a makefile will automate the bulk of the process, but there are some minimum packages needed for Nauta).

  1. Populate or create Nauta's inventory file to define where your master and worker nodes are, and how to access them. To create the Inventory file:

    • Copy the Inventory file example information: Inventory File Information.

    • Modify the newly created Inventory file to suit your needs.

  2. Modify Nauta's configuration file to define your proxy, network quirks and filesystem preferences. To create the configuration file:

  3. Run the installation script (see Installation Process for more information).

This process does the following:

  • Creates a Kubernetes cluster, all the Docker files you need to run Tensorflow, Jupyter, Tensorboard, and Horovod.

  • Installs the Nauta server-side application on your new Kubernetes cluster, and starts the system running.

Completing all of the above takes some time. We are working on streamlining the process.

Removing Nauta

Should you need to delete/remove Nauta, refer to: Installation Process page for more details.

Document Flow

This guide consists of the following main topics, in order. Start here to move through the correct sequence of events.

To read Intel Terms and Conditions check this document.

This user guide is subject to CC-BY-ND 4.0.