Exponential Backoff + Jitter

Problem

10,000 clients lose connection simultaneously.

If all retry one second later:
10,000 failures
      ↓ 1 second
10,000 retries

the recovering service gets overwhelmed again.

Solution — Backoff + Jitter

Clients progressively wait longer and add randomness. Jitter is a small random variation in retry timing. It prevents many clients from retrying simultaneously and overloading a recovering service.

1.2s → 2.7s → 5.1s → 9.8s

Retries spread across time instead of arriving together.

Core idea: backoff and jitter prevent retries from amplifying an outage.