Quorum

A quorum is the number of replicas that must respond before an operation is considered complete. Rather than waiting for every copy — slow, and unavailable whenever one replica is down — or accepting a single copy — fast, and possibly stale — a quorum waits for a defined subset.

The useful property is arithmetic. With N replicas, if a write must reach W of them and a read must consult R of them, then choosing W + R greater than N guarantees that the read set and the write set overlap in at least one replica. That overlapping replica has the latest write, so the read can find it. Nothing about the network needs to be assumed for this to hold; it follows from counting.

This turns consistency into a dial rather than a fixed property. Raising W makes writes slower and more durable; raising R makes reads slower and fresher. Many systems expose these per operation, so one query can demand the overlap while another accepts a single replica for speed.

Two limits are worth stating plainly, because quorums are easy to over-trust. A quorum bounds latency by the slowest replica that must answer, so a distant or degraded replica in the required set slows every operation. And overlap guarantees that the newest value is available to the reader; it does not by itself make concurrent operations appear in a single global order. Systems offering linearizable operations on top of quorums add protocol to get there.

Quorums also appear in consensus protocols, where a majority must agree before a decision is committed — the same counting argument, applied to agreement rather than to data freshness.

How the dial interacts with availability during a partition is worked through in Replication, Consistency, and Time.