Jitter

The core role of jitter is to break the thundering herd phenomenon in which multiple clients retry in sync due to identical backoff curves, and it can significantly reduce retry storms in high-concurrency scenarios.

The ±10% jitter causes all clients that failed at 0 seconds to wait 450-550 milliseconds, so that 80 retries arrive crowded into the same 100-millisecond window, forming waves and failing to effectively spread out retry timing.

Jitter strategies spread out retry moments by introducing a random component; common algorithms include full jitter, equal jitter, and decorrelated jitter; adding jitter does not lengthen the average retry time, yet significantly reduces the peak pressure of synchronized retries.

Exponential backoff means that each retry increases the wait time by a multiple (such as 1s, 2s, 4s, 8s); jitter is adding a random amount to the calculated wait time to prevent a large number of clients from retrying simultaneously at the same moment due to using the same backoff curve, which would create a new traffic pulse; common approaches include 'full jitter' and 'equal jitter.'

Jitter introduces a random amount into the backoff interval to prevent large numbers of clients from retrying synchronously at the same moment and causing a thundering herd.