The Overlap Rule
In a replicated store with N copies, a write contacts W replicas and a read contacts R replicas. The key inequality is:
- If W plus R is greater than N, the read set and write set must share at least one replica.
- That shared replica holds the latest write, so the read can find it.
This gives strong consistency through overlap rather than through a single master.
Tuning the Numbers
- W equals N and R equals one makes reads cheap and writes expensive.
- W equals one and R equals N flips that trade.
- W plus R equals N plus one is the smallest overlap that stays consistent, often W and R both set near half of N plus one.
When Overlap Is Lost
If you pick W plus R not greater than N for speed, read and write sets can miss each other, so a read may return a stale value. That is a deliberate availability and latency choice, not a bug, but you must label the data as eventually consistent.
Key idea
When W plus R exceeds N the read and write sets overlap on a replica holding the newest value, giving consistency you can tune for latency.