One database, three continents
An airline sells seats on flight UA88 from three data centers: Virginia, Frankfurt and Singapore. Each one keeps a full copy of the database. Each copy is a replica.
Why three? If a data center loses power, the other two still have every booking. And each customer talks to the replica nearest to them: Alice in the US, Bob in Europe, Carol in Asia. Staying on one continent takes a millisecond or two instead of a couple of hundred.
Watch Alice book a seat. Virginia updates its copy (seats_left drops and the version goes from v7 to v8), answers her, then sends the change to the other two. That's replication.
The numbers on the links are typical round-trip times between these cloud regions. The animation runs about 10× slower than real life so you can follow every message.
CREATE KEYSPACE booking WITH replication = {'class': 'NetworkTopologyStrategy','virginia': 1, 'frankfurt': 1, 'singapore': 1};
Cassandra: keep one full copy of every row in each data center. The data-center names come from your cluster’s own configuration.