Course contentsShow
← Computer Science
Lesson 2593 of 2,87258. Distributed Systems BasicsFree lesson

Node Failures: Crash and Byzantine

Distinguish between crash failures (nodes stop) and Byzantine failures (nodes behave arbitrarily or maliciously).

Node Failures: Crash and Byzantine

You'll learn the two fundamental models of how nodes fail in distributed systems: crash failures and Byzantine failures.

The Two Failure Models

When a node in a distributed system fails, it can do so in dramatically different ways. Understanding these failure modes is critical for designing resilient systems.

Crash failures are the simpler, "well-behaved" failure model. When a node experiences a crash failure, it simply stops working—like a computer that loses power. It stops sending messages, stops processing requests, and essentially vanishes from the system. Crucially, it doesn't send incorrect or malicious data; it just goes silent. Think of it like a colleague who doesn't show up to work: they're not actively causing problems, they're just absent.

Byzantine failures are far more challenging. Named after the "Byzantine Generals Problem," these failures allow a node to behave arbitrarily. A Byzantine node might:

  • Send corrupted or contradictory data
  • Lie about its state
  • Collude with other faulty nodes
  • Selectively respond to some nodes but not others
  • Appear to work correctly sometimes and fail at others

Imagine a colleague who not only doesn't do their work but actively sabotages projects, lies in meetings, and gives different people contradictory information. That's Byzantine behavior.

Why It Matters

Most distributed systems are designed to handle crash failures—they're predictable and detectable. Byzantine failures require much more sophisticated (and expensive) protocols like Byzantine Fault Tolerance (BFT), which typically need more than two-thirds of nodes to be honest to reach consensus.

Systems handling cryptocurrency or military communications must defend against Byzantine failures. Most internal corporate systems only worry about crash failures.

Key Takeaway: Crash failures mean nodes stop cleanly; Byzantine failures mean nodes behave maliciously or arbitrarily—requiring fundamentally different defensive strategies.