Summary

In this chapter, we discussed the various managed deployment methods available when using Amazon SageMaker. We talked about the suitability of the different deployment/inference methods for different use case types. We showed examples of how we can do batch inference and deploy real-time and asynchronous endpoints. We also discussed how SageMaker can be configured to automatically scale both up and down, and how SageMaker ensures that in case of an outage, our endpoints are deployed to multiple availability zones. We also touched upon the various blue/green deployment methodologies available with Amazon SageMaker, in order to update our endpoints in production.

In a lot of real-world scenarios, we do not have high-performance clusters of instances available for carrying out inference on new and unseen data in real time. For such applications, we need to use edge computing devices. These devices often have limitations on compute power, memory, connectivity, and bandwidth...