Quorum reads and writes overlap

With N copies of each item, a write waits for W of them and a read consults R. When R plus W is greater than N, the set written and the set read must share a copy, so every read reaches at least one holding the latest write. Three copies, written to two and read from two, is the usual setting.

The cost, in Gorton's account, is that reads and writes fail when no quorum is reachable, as for a client on the minority side of a partition.

It is a looser relative of consensus. The overlap finds the newest value but does not order concurrent writes to the same key. Last writer wins, the common rule, silently discards one, and clock drift makes its timestamp order arbitrary: Reconciling divergent writes is a product decision.

The overlap holds only while the quorum is strict. Cassandra, Riak and other Dynamo-style stores offer sloppy quorums, parking writes on whichever nodes are reachable and handing them off later. Writes stay available but reads can be stale, a CAP choice that is usually a configuration switch. A larger W also lengthens every write, which across regions is synchronous replication by another name.