Questions: Consistency Models in Distributed Systems

Question 1

Two consistency models are commonly confused. Which one is strictly stronger: linearizability or sequential consistency?

• A. Sequential consistency, because it allows more orderings
• B. Linearizability, because it additionally requires real-time ordering of operations
• C. They are equivalent — both require a single global order
• D. Neither; they are incomparable models

Answer: (B)

Linearizability (also called atomic consistency) requires that operations appear to execute instantaneously at some point between their invocation and completion, preserving real-time order. Sequential consistency only requires that all nodes observe operations in the same order, but that order does not need to match wall-clock time. Linearizability is therefore strictly stronger.

Question 2

Under eventual consistency, if a client writes a value to one replica and then immediately reads from a different replica, the read is very likely to return the new value.

• T. True
• F. False

Answer: False

Eventual consistency only guarantees that replicas will *eventually* converge to the same value if no new writes occur — it makes no promise about when. A read immediately after a write may return a stale value from a replica that has not yet received the update. This is the fundamental tradeoff: lower latency and higher availability at the cost of temporarily stale reads.

Question 3

Why do distributed systems designers often choose eventual consistency over linearizability, even when they would prefer strong consistency semantics?

Model answer: The CAP theorem shows that a distributed system cannot simultaneously guarantee consistency, availability, and partition tolerance. Linearizability requires coordination (e.g., quorum reads/writes) that adds latency and becomes unavailable during network partitions. Eventual consistency allows systems to remain available and low-latency by deferring convergence.

This is the core design tension in distributed systems. Systems like Amazon DynamoDB and Apache Cassandra chose eventual consistency to achieve high availability across data centers. Systems like Google Spanner accept higher latency to provide linearizability. The choice depends on whether the application can tolerate stale reads.