Exponential Backoff Planner

Exponential backoff is the policy of waiting longer between each attempt: 100 ms, then 200, then 400. The idea is giving the failing service room instead of hammering it. Fill in the base delay, the multiplier, the cap and the number of attempts, and the page builds the full schedule — each individual wait plus the total in the best, expected and worst case.

The sum almost everyone gets wrong is the number of waits. Six attempts means five waits, not six: the first attempt happens immediately, and only after it fails does anything wait. Counting one wait too many overstates the total by the largest one — with a multiplier of 2, the final wait is worth nearly as much as all the previous ones combined.

The choice of jitter strategy matters more than it looks. Without jitter, a thousand clients that failed at the same instant come back at the same instant, and the surge knocks over the service that was getting back up. Full jitter draws between zero and the nominal delay: it spreads the load best and halves the expected time without raising the worst case. Equal jitter keeps half fixed, guaranteeing a minimum wait. And decorrelated grows from the previous drawn wait rather than the nominal one, spreading further at the cost of predictability.