Retransmission requires deciding when an unacknowledged transmission should be considered lost. A fixed timeout performs poorly because network delay varies.

A sender can measure round-trip time (RTT) from transmission to acknowledgement and maintain a smoothed estimate together with a measure of variation. Its retransmission timeout should be longer than a typical RTT but still react promptly to genuine loss.

If the timeout is too short, delayed packets trigger unnecessary retransmissions and add congestion. If it is too long, recovery from real loss is slow.

For example, a path whose RTT fluctuates around 80 ms should not use a 20 ms timeout merely because one fast sample was observed.

Adaptive timers illustrate a general network principle: protocol decisions must account for uncertain and changing delay rather than assuming deterministic timing.