Author: Howard M. Cohen
Were you to ask Microsoft why you should be using Azure to power your artificial intelligence (AI) based applications and workloads, they would respond:
“Successful transformation with AI starts with a powerful, secure, and adaptive infrastructure strategy. And as you evolve, you need a cloud platform that adapts and scales with your needs.
Azure is that platform, providing the optimal environment for integrating your applications and data so that you can start innovating with AI.
As you design, deploy, and manage your environment and workloads on Azure, you have access to best practices and industry-leading technical guidance to help you accelerate your AI adoption and achieve your business goals.”
Getting Started is Half the Battle
Perhaps the best way to think about Azure, to visualize your plan from the start, is to look at it as an environment. Far more than an architecture, a programming construct, or a feature, AI exists in an environment that connects it to all the data, all the elements, and all the resources it needs to perform.
Getting an AI application running in such an Azure environment requires several key steps.
Create the Foundation
The building of an AI-ready environment in Azure begins with the design for the resource hierarchy and the provisioning of the core infrastructure. You need to set up a secure environment with plenty of room for scaling. Take care to align this environment with your data governance and other operational requirements. This will facilitate far more efficient deployments and ongoing management of your AI workloads.
Azure Landing Zones
For many, Azure Landing Zones which are set up following key design principles across the fundamental design areas of Azure to accommodate all application portfolios and enable application migration, modernization, and innovation at scale. In many ways, this provides a pre-designed template or starting point on which to develop your new Azure environment.
Azure Landing Zones use subscriptions to isolate and scale application resources and platform resources. Application Landing Zones are the subscriptions used to host any application. Platform Landing Zones provide the required shared services such as identity, connectivity, and management.
AI Governance
It is often said that everything begins with policy, and that is certainly the case when building your Azure AI environment. The resources required to run the environment must be carefully organized and policies must be established regarding the acceptable use, compliance, security, and costs. When done properly, governance prevents unauthorized access, manages risk, and assures efficiency.
It is here that you will establish management groups for both internal and external facing AI workloads, assuring they will be kept separate. You will create and apply policies for each of these workload types.
AI Networking
AI requires highly efficient, high-speed communications to move all that data to and from where it needs to be, and it must always be transported securely and reliably.
It is recommended that you activate Azure Direct Denial of Service (DDoS) Protection for all internet-facing AI workloads to protect those workloads from possible disruption and/or downtime that can be caused when bad actors attempt to overload your network with requests.
There are several other Azure utilities you may wish to add for greater protection, including Azure Bastion, Azure ExpressRoute with FastPath, Azure VPN Gateway, Azure Firewall, and Azure DNS.
Assuring AI Reliability
Reliability is the product of a series of provisions made to assure consistent performance, compliance, and all-important availability.
It is recommended that you provision AI hosting among several Azure regions to provide redundancy and high availability. This assures faster failover and recovery from potential disruptions. Before provisioning be sure to assure availability of the services you will need to support your AI applications. Also become familiar with any regional quotas or capacities to assure they can support your application and workloads without fail.
Also research and plan your data storage requirements and assure availability of resources where you’ll most need them.
Although it is true that Azure incorporates redundancy by its fundamental design, you must yourself assure business continuity by replicating critical assets such as fine-tuned models, retrieval augmented generation (RAG) data and other trained models and datasets in secondary regions.
There’s Much More… But…
This being only a partial inventory of all the considerations and processes which must be executed before you can deploy your AI-based applications on Azure, you may see deployment of AI-based applications on Azure as a very challenging project. And there’s no question that it is.
This is not, however, a journey you must go on alone. When working with a partner that is thoroughly familiar and fully prepared to manage your AI workloads, you will find yourself able to concentrate fully on what each application does for you and your organization.
Idenxt is just such a partner. The technology team that designed and built Idenxt has been working with Azure for decades and is ready to help you migrate and manage your AI-based applications.
To learn more, contact us here.