A system has N=5 replicas with write quorum W=3 and read quorum R=3. A client reads from 3 replicas. Can it be guaranteed that at least one of those replicas has the most recently written value?
ANo — it depends on which 3 replicas are contacted; if none of the 3 that acknowledged the write are in the read set, the latest value is missed
BYes — because W+R=6 > N=5, any set of 3 read replicas must overlap with any set of 3 write replicas by the pigeonhole principle
COnly if the system uses a leader that tracks which replicas are up to date
DOnly if all 5 replicas are available; if any replica is down the guarantee fails
This is the mathematical core of quorum systems. If W+R > N, then the write set (W replicas) and any read set (R replicas) together account for more than N total slots. Since there are only N replicas, the two sets must share at least one member. That shared member has seen the latest write. With W=3, R=3, N=5: 3+3=6 > 5, so at least one replica in any read quorum of 3 also received the latest write quorum of 3. This is a guarantee for *any* choice of 3 read replicas — not just a lucky one.
Question 2 Multiple Choice
A designer sets W=1 and R=N for an N-replica system. What is the effect on read and write performance, and is this less reliable than a majority quorum?
AWrites become faster, reads become slower, and the system is less reliable because only one copy holds the latest data
BWrites are very fast (only one acknowledgment needed), reads are slow (must contact all replicas), but consistency is still guaranteed as long as W+R > N
CThis configuration violates W+R > N, so it is invalid and provides no consistency guarantee
DBoth reads and writes are slower because contacting all replicas for reads adds latency that offsets write gains
W=1, R=N: 1+N = N+1 > N, so W+R > N is satisfied. Every write is acknowledged by just one replica (fast writes), but every read must contact all N replicas (slow reads) — and the highest-version response is the latest write. This is a legitimate configuration with its own tradeoff profile. It is not less reliable in terms of correctness, but it is less available for reads during failures: if any replica is down, R=N cannot be satisfied. Reliability and performance tradeoffs come from where you set W and R, not from violating the quorum invariant.
Question 3 True / False
Meeting the quorum invariant W+R > N guarantees that a read will typically return the most recently written value without any additional protocol mechanisms.
TTrue
FFalse
Answer: False
The quorum invariant guarantees that at least one replica in the read set *has* the latest value — but it does not automatically return it to the client. If the client simply takes the first response, a slow replica might reply with stale data first. Version numbers, timestamps, or vector clocks must be attached to each response so the client can identify which is freshest. Some systems also use read repair: on discovering stale replicas in the quorum, the latest value is pushed back to them. The quorum overlap is the mathematical prerequisite; the protocol on top determines whether that freshness guarantee is actually realized.
Question 4 True / False
In quorum-based replication, increasing the write quorum W usually improves read performance.
TTrue
FFalse
Answer: False
Increasing W means more replicas hold the latest write, which allows R to decrease while still satisfying W+R > N — so yes, it can enable faster reads. But increasing W directly *slows writes*, since more replicas must acknowledge each write. There is no free lunch: the sum W+R must exceed N, so making writes heavier (higher W) creates room to make reads lighter (lower R), and vice versa. If W is increased while R is held constant, there is no improvement in read performance — just slower writes with more replicas storing the latest data.
Question 5 Short Answer
Explain in your own words why the invariant W+R > N guarantees that any read quorum must contain at least one replica that has seen the most recent write.
Think about your answer, then reveal below.
Model answer: If W replicas must acknowledge every write, those W replicas form the 'write set.' If R replicas must respond to every read, those R replicas form the 'read set.' Since there are only N replicas total, and W+R > N, the write set and read set together claim more than N slots among N replicas — so they must share at least one replica by the pigeonhole principle. That shared replica received the latest write and is in the read set, so the reader can always find the latest version by taking the highest-version response across the R replies.
This is the only theorem needed to understand quorum systems. Everything else — version numbers, read repair, tunable W and R — is engineering built on top of this guarantee. A common misconception is that quorums require luck or that freshness depends on which nodes happen to respond. The invariant eliminates that luck: *any* R nodes will include at least one write-quorum member, regardless of which ones respond first.