3

Architectural and Cost Optimization Considerations

In this chapter, we are going to focus on some of the other services you will likely need in most, if not all, deployments of Cognitive Services within Azure. We will discuss Ocean Smart evaluated options and demonstrate how they fit into an overall architecture to build a full solution. After describing these services, we will discuss how Ocean Smart was able to optimize its costs while building the complete solution. We will also cover what considerations should be made as you progress from PoC to Pilot, to user testing, and finally to production. This can be very complicated as it requires significant attention to detail and constantly keeping an eye on deployments to minimize costs and follow best practices.

This chapter will cover the following topics:

• Exploring core Azure services and costs
• Estimating the costs for a proof-of-concept solution, including all services
• Understanding data orchestration for...

Exploring core Azure services and costs

In this section, we are going to look at what services Ocean Smart used when building solutions in Azure. When examining some of the architectures provided by the Azure documentation, it is generally assumed that the data that is being used for the solution has already been uploaded into Azure, except for streaming data examples. Later in this chapter, we will look at some of the options you have for getting your data into Azure for batch loading and streaming.

When considering costs, as described in each of this chapter's sections, we will always provide a balance of costs and services for optimal configurations. As it goes, the better the service, the higher the cost. Only you can determine what your business requirements are, with help from other stakeholders, to ensure an optimal mix of services and costs. It is important to note that service-level agreements (SLAs) do not guarantee service – they simply provide a mechanism...

Estimating the costs for a proof-of-concept solution, including all services

To build relatively accurate costing models, Ocean Smart needed certain estimates to be made based on logic and some guesswork. When you're building your costing models, be mindful that it is very difficult to get exact costs initially. As Ocean Smart was able to bring the solution that was developed into a production state, they started to see how long it took to move data, how much data was being stored to maintain the model, how much compute was required, and how long it would take to execute model training and retraining. In turn, they began to build a basis to add more compute that was required to meet internal SLAs for processing time. The remainder of this section will go through the decisions they needed to make when it came to building an estimate for a Form Recognizer proof of concept solution in Azure. The prices and tiers of each of the components are built in the East US 2 region of Azure...

Understanding data orchestration for loading data into Azure

To get data into Azure, Ocean Smart had a multitude of choices for orchestration, with a variety of suggested uses. These tools can use either a command-line interface (CLI) or graphical user interface (GUI) and have various capabilities for automating, scheduling, and triggering actions. These tools can also be local to a user's workstation for moving the data or purely reside in the Azure portal. First, we will start with the workstation-based tools:

• AzCopy: AzCopy is a command-line tool that is installed on a user workstation or server within an organization. It is intended to be used to send files into an Azure storage account from the local computer. Using this tool requires secure connectivity using TLS to connect and send files to the storage account in the desired folder. The account that's being used can be authenticated using Azure Active Directory or using a Shared Access Signature (SAS) key....

Summary

This chapter covered the essential information that Ocean Smart used to understand the common Azure services that are used with cognitive services deployments. We also explored the most important things we must consider when deploying the services successfully at optimal costs, as well as their corresponding SLAs. We also went through a simple pricing exercise by comparing service tiers, looking at the services that are required to build a Form Recognizer PoC, and exporting the pricing calculator. These estimates will help you build the ROI and TCO models that we discussed in Chapter 2, Why Azure Cognitive Services?. As we diver deeper into this book and the build solutions for the specified use cases, we will look closer at the modeling costs for the solutions. Finally, we wrapped up this chapter by providing an overview of some data movement tools that can help you get your data moved to Azure efficiently for your requirements.

As we discussed earlier in this book, SLA...