In-memory data store in the Ballerina: implementation, limitations, and scaling

I recently came across the in-memory datastore support of bal persist feature. Regarding this, could you provide the following details?

• What is the underlying implementation of this?
• How well does it scale?
• What are the limitations?

To store data in in-memory, we use Ballerina tables underneath and use generated clients to access the tables. There is a Ballerina table created for each entity defined in the model. in-memory data store supports all Ballerina types, but this is a volatile memory.