Overall impression

We run two sets of tests:

• "Regular" test, with an index only over the objects IDs;
• "Unique field" test, with an index over object IDs and a "Name" UTF-8 property.

The performance scaling, when it comes to mORMot internal process (serialization, ORM, REST, in-memory tables) and SQLite3 engine, is linear.
Existing bottlenecks (like SQLite3 hard-drive on disk synchronization) persists, which does make sense.

Oracle performance is similar for individual requests, since the most process is spent in the remote access over a network; but array DML insertion is still very efficient, and CPU % use is very low.
TObjectList / TSQLRestServerStaticInMemoryExternal process speed is amazing - and the new CPU seems to like it!

ODBC process is still slower than direct access to Oracle client, and could be enhanced - especially by implementing array binding of parameters for BATCH process.
But no regression has been observed, just some small fixes discovered and committed.

Regular test

First of all, the "regular test", with UNIK conditional not defined.
It is similar to those previous tests.

Insertion speed

Read speed

Test with unique field

Then, we defined UNIK conditional, to compare with this other benchmark result.

Insertion speed

Read speed

Feedback is welcome on our forum!