The Last Instance Type

A write-heavy primary database already runs on the largest instance type the provider offers and is still saturated on write throughput. The application tier itself is stateless and scales fine. What is the honest next move?

Partition writes across multiple primaries, or cut write volume, since vertical scaling has run out of machine

Enlarge the application's connection pool so more writes can be in flight against the primary at once

Put a larger read cache in front of the database so most of the traffic is absorbed before it reaches the primary's write path

Add read replicas so the write stream is spread over several nodes instead of being applied by the primary alone