A.1.5 Who owns and manages the VMs on which embedded Python execution is downloaded?

Is the VM to which the embedded Python execution workload is offloaded, owned and managed by OML (service-owned )? Or is it under the ownership of the database itself (customer-owned )?

For Oracle Machine Learning, each tenant has its own VM managed by Autonomous AI Database. Embedded execution does not run on customer-owned VM. Multiple VMs might belong to a tenant but no VM can be shared across different tenants. Through the service levels - Low, Medium, High and GPU settings, the different sized containers can be provisioned respectively. There is no dynamic resource adjustment in place at runtime.

A tenant may provision one or more databases. Oracle Autonomous AI Database Serverless maintains a pool of VMs. Each embedded execution is processed in a container, which runs on a VM. When we need a VM to run a container, we request a VM from Oracle Autonomous AI Database Serverless. The VM is released when no more containers are running. Currently, a VM only serves one database at a time. All database users share this VM). If a tenant has 2 databases, each database will have its own VM.

Note:

Oracle Autonomous AI Database Serverless does not support dynamic resource adjustment for the containers.