The title is misleading and exaggerating the scale of the test. The test is essentially having a billion users in the database and subsequently load testing for merely 7,500 concurrent requests. The traffic assumption is as simple as 150 rps peak traffic discovered for 20mil users, so for 1bil users (20mil x 50) it would be 7,500 rps (150 x 50). Database insert rate of 2,000 users/sec means the approach is elementary and inefficient. It’s okay if this is the first real performance test you’ve done as it takes time to articulate traffic modelling and develop meaningful load tests for very large scale systems.