[ddorian43]

Imagine single core single threaded design. You send 2 requests for 1 row each.

First request the row needs to be read from disk HDD. It takes 2ms.

Second request, the row is already in ram, it takes microseconds but still has to wait for the first request to finish.

Threads have overhead when having a lot of concurrency (thousands/millions requests/second).

For extreme async, see seastar-framework and scylladb design.

TLDR: high concurrency, low overhead etc.