Course contentsShow
← Computer Science
Lesson 2611 of 2,87258. Distributed Systems BasicsFree lesson

Strong Consistency (Linearizability)

Learn linearizability: operations appear to execute atomically in a single total order matching real-time.

Strong Consistency (Linearizability)

What You'll Learn

You'll understand linearizability, the strongest consistency model where every operation appears to execute atomically at a single instant in real-time order.

The Core Idea

Imagine a single-file line at a coffee shop. Each person orders, gets their drink, and leaves—one at a time. No matter who's watching, everyone agrees on the order things happened. Linearizability makes a distributed system behave exactly like this: as if there's just one copy of the data, and every operation happens instantly at some point between when it started and when it finished.

In a linearizable system:

  1. Operations appear atomic—they take effect instantaneously at some point during their execution
  2. All clients see operations in a single, total order
  3. This order respects real-time: if operation A completes before operation B starts, then A must appear before B in the order

Think of it as "one at a time, right now." If you write a value and the write completes, every subsequent read must see that value (or a newer one). No exceptions, no delays, no stale reads.

Why It Matters

Linearizability is the gold standard because it makes distributed systems act like single-machine programs. You can reason about them intuitively—what happens "happens," and everyone agrees. This eliminates entire classes of subtle bugs.

The trade-off? It's expensive. Achieving linearizability typically requires coordination across nodes, which means higher latency and reduced availability during partitions (recall the CAP theorem: you sacrifice availability to maintain consistency during network splits).

Key Takeaway: Linearizability guarantees operations appear to execute atomically in a single global order that matches real-time causality, making distributed systems behave like centralized ones at the cost of performance and availability.