How computers learn to wait when the world gets too loud

Countries & Cities · Published 12 October 2026 Swipe more like this →

In short: When a network becomes congested, thousands of devices might try to speak at once, causing a digital traffic jam. To solve this, systems use an algorithm that forces them to wait longer after every failure. It is a mathematical dance of retreat and recovery that keeps our modern internet from collapsing.

Exponential backoff is a closed-loop control system designed to find an acceptable rate of operation by reacting to adverse events like packet loss, network collisions, or service errors. Instead of retrying an action immediately—which would likely cause another failure—the system waits for a duration that increases multiplicatively. For instance, if a mobile app fails to connect, it might wait 1 second, then 2, then 4, and so on. This is known as a binary exponential backoff when the multiplicative factor is 2.

The logic is essential for shared channels where multiple senders communicate over a single frequency. If two senders transmit simultaneously, a collision occurs. If they both retried at the exact same fixed interval, they would collide again indefinitely. To prevent this, the algorithm introduces randomness. In a binary exponential backoff, after a collision, each sender chooses a random delay within a range that grows exponentially. For example, after the first collision, a sender waits between 0 and 1 slot times; after the third, it waits between 0 and 7 slot times. This increases the probability of success while managing the trade-off with latency.

The history of this logic dates back to the early 1970s. Norman Abramson explored users on different islands sharing a single radio channel to access a main computer at the University of Hawaii. Later, Larry Roberts and Simon Lam developed models for slotted ALOHA, a protocol where users choose random time slots. Lam’s research used Markov chain models to prove that while these systems can be inherently unstable under certain conditions, increasing the range of possible backoff slots (K) can increase the maximum number of users a channel can accommodate. Today, these principles govern everything from Ethernet networks to web service rate limiting.

General information, not professional advice. Spotted an error? Flag the card or see corrections and complaints.

Intellect Plus is a visual educational discovery platform for exploring fascinating facts and ideas. Brain Food, Not Brain Rot. What is Intellect Plus?

Related

• How the Transmission Control Protocol keeps the modern internet from falling apart
• How the internet keeps perfect time across billions of devices simultaneously
• How ancient algorithms prevent the modern internet from collapsing
• The Internet Protocol is designed to be unreliable, and that is its greatest strength.
• Why the Internet's fastest protocol is intentionally designed to be unreliable
• The Berlin TV Tower was a communist power symbol that became a global icon.
• The mechanism that makes firing federal policy-makers easier
• London's most famous empty pedestal was never meant to be a blank canvas
• The world's only binational airport sits entirely in France but functions as two countries.
• How a 42-metre monument to freedom survived decades of Soviet occupation in Riga
• The invisible structure that carries the modern internet
• Why did a 630-foot stainless steel arch become a lightning rod for fascist accusations?

More in Famous landmarks · More in Countries & Cities · Latest · A–Z · All topics