The core idea

A fast shared layer in front of slow stores.

At scale, a distributed in-memory cache sits between your servers and the database, absorbing reads that would otherwise crush it.