Intel® Optimized Cloud Recipes (OCRs)

© Copyright 2023, Intel Corporation

Overview

Intel® Optimized Cloud Recipes (OCRs) are:

1. A combination of documentation(markdown) and scripted automation.
2. Developed to enable/optimize technologies on Intel® Architecture (IA).
3. Scripts that depend on existing Intel® Tuning Guides that were generated from testing and benchmarking exercises.
4. Currently focused on enabling Infrastructure-as-a-Service (IaaS) Virtual Machines (VMs) on public cloud providers (today, on-prem hardware configuration is not supported).
5. Written in Ansible or PowerShell, meant to automate operating system and above configuration.
6. Ready to be Integrated into Intel® Cloud Optimization Modules for HashiCorp Terraform using Cloud Cloud-Init integration. See GCP Module example).
7. Can also be used directly on their own(see usage on each recipe).

Currently out-of-scope for OCRs:

1. Physical on-prem hardware configuration.
2. Defining tuning parameters/knobs and values that are required to optimize a workload.
3. Benchmarking.

Note: OCRs focus on the enablement of Intel Technologies and often will include a demo but are not meant to be a benchmarking tool.

Intel is working on a separate benchmarking repository called Workload Services Framework that will consume these recipes and provide benchmarking capabilities.

Value

The primary goal of OCRs is to simplify and accelerate the enablement of Intel® Technologies on public cloud VMs.

Examples of Recipes

Some examples of technologies that can be enabled and optimized are:

• OneAPI AI Training and Inference with Intel® Advanced Matrix Extensions (Intel® AMX)
• Encryption/Decryption with Intel Quick Assist Technology (Intel QAT)
• MySQL Open Source Database
• PostgreSQL Open Source Database

Intel® Cloud Optimization Modules for HashiCorp, the OCRs companion

The Intel® Cloud Optimization Modules for HashiCorp Terraform enable the deployment of optimized instances on public cloud. The modules are available on the Terraform Registry.

The Modules can be easily integrated with these Recipes, see Usage below.

Usage

Recipes are located in the ./recipes folder. Each recipe has its own folder and contains a README.md file with instructions on how to use it.

There are two primary ways to use these Recipes:

1. Via a Cloud-init Integration with Intel® Cloud Optimization Modules for HashiCorp Terraform
2. By running Ansible via the command line

Option 1 - Integration with Intel® Cloud Optimization Modules for HashiCorp Terraform via Cloud-Init

Overview:

1. The following demonstrates how to use the ai-pytorch-amx-ubuntu recipe with the Intel GCP module for Terraform.
2. It uses the Terraform user_data argument that enables the execution on the cloud_init.yml file.
3. The cloud_init.yml file calls the Ansible playbook that installs the recipe by calling the recipe.yml file directly from Github.

GCP Module example link

Example:

# main.tf Terraform file that deploys GCP VM via the GCP VM Module

# Cloud-init file that calls Ansible playbook, see 'user_data' below
data "template_file" "user_data" {
  template = file("./cloud_init.yml")
}

# GCP VM Module
module "linux_vm" {
  source              = "intel/gcp-vm/intel"
  project             = "123456789"
  boot_image_project  = "ubuntu-os-cloud"
  boot_image_family   = "ubuntu-2204-lts"
  name                = "gcp-linux-with-aikit"
  zone                = "us-central1-a" 
  access_config = [{
    nat_ip                 = null
    public_ptr_domain_name = null
    network_tier           = "PREMIUM"
  }, ]
  # Integration between Terraform and Cloud-init
  user_data    = data.template_file.user_data.rendered 
}

# Cloud-init file that calls Ansible playbook
package_update: true
package_upgrade: true

package:
  - git

ansible:
  install_method: distro
  package_name: ansible
  pull:
    url: "https://github.com/intel/optimized-cloud-recipes.git"
    playbook_name: "recipes/ai-pytorch-amx-ubuntu/recipe.yml"

Option 2 - Running Ansible via the Operating System command line

Follow Ansible installation procedures for your operating system

https://docs.ansible.com/ansible/latest/installation_guide/installation_distros.html

For example, on Ubuntu by using ansible-pull, Ansible can run directly on the host:

# Install Git 
sudo apt install git -y

# Install Ansible Key
sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 93C4A3FD7BB9C367
sudo apt update

# Install Ansible
sudo apt install ansible -y

#Run ansible-pull
sudo ansible-pull -vv -U https://github.com/intel/optimized-cloud-recipes.git recipes/ai-pytorch-amx-ubuntu/recipe.yml

# Logs at 'tail -f 10 /var/ansible-log' & 'tail -f 10 /var/log/dpkg.log'

Repo Structure

The repository structure is very simple. All recipes will be located under the ./recipes folder.

The recipes folders are structured as <category>-<workload name>-<intel tech enabled>-<OS>

Categories are: ai, compute, data, media, ml, networking

Examples:

• ai-pytorch-amx-ubuntu
• networking-java-qat-ubuntu
• database-mysql-avx512-ubuntu
• database-mssql-qat-windows

Integrations with Intel® Cloud Optimization Modules for HashiCorp Terraform

See Usage -> Option 1 above

A diagram showcasing the integration is coming!

Development Environment

We recommend using a Linux environment. If on Windows, we recommend using WSL2.

You will only need git and Ansible installed on your development environment. Follow Ansible installation procedures for your operating system. https://docs.ansible.com/ansible/latest/installation_guide/installation_distros.html

For example, on Ubuntu:

sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 93C4A3FD7BB9C367
sudo apt update
sudo apt install git -y
sudo apt install ansible -y

Note: If you are an Intel® employee, reach out to the Repository Owners. We have additional environment configuration information that can help you.

How to contribute

Please see CONTRIBUTING.md

We are glad to review and accept contributions.

Considerations

Today, the scope is limited to enabling/optimizing VMs on public cloud providers.

The OCRs are currently not managing on-prem physical hardware configurations.