The image shows the basic deployment of machine learning models close to data sources and production systems. It shows both a customer on-premises implementation and an Oracle Cloud implementation.

The on-premises implementation contains production machinery, which passes runtime IoT data to a machine language execution compute instance, which then passes it to custome premises equipment (CPE). From there, data moves over a site-to-site VPN to Oracle Cloud.

Within Oracle Cloud, a dynamic routing gateway (DRG), passes incoming IoT production data in a virtual cloud network (VCN), from which a service gateway that routes traffic to an Oracle Streaming Service instance outside the VCN. The streaming service passes data through an Oracle Functions instance, and then into an autonomous data warehouse. An Oracle Data Science instance fetches data from the ADW and then passes it to Oracle Object Storage.

The model updates are then passed back through the service gateway to the VCN and then back to the customer on-premises implementationthrough the DRG, the site-to-site VPN, the CPE, and terminated at the ML execution compute instance.