Leaderless replication
Replication with no single leader: a client sends each write to several replicas and reads from several, and a quorum decides. Amazon's Dynamo made the design well known, and Cassandra uses it. It keeps writing when some replicas fail, at the cost of reads that may be stale unless R + W > N.
Data
In Quorum#
The Leaderless replication Research (after Sharding) unlocks the Wide-column Store. Each key lives on N replicas (1 to 5, default 3), a read asks R of them, and a write waits for W (default 1 and 1). With R + W > N every read sees the latest write; with less, some reads come back stale. A write costs N replica operations and a read costs R, out of 30,000 per second per size step, and if an Instance failure takes one replica, quorums that need every replica fail.