latency

term

networking

The time between sending something and it arriving - distinct from bandwidth, which is how much arrives per second.

The two get confused constantly, and the confusion is expensive: adding capacity does nothing for a slow application whose problem is round trips. Latency has four parts - propagation, which is physics and cannot be bought down; serialisation, which shrinks as links get faster; queueing, which grows with load and is where most surprises live; and processing. Distance sets a floor no product can beat, which is why a chatty protocol crossing an ocean is slow no matter how much bandwidth is bought, and why moving the work closer is the only real fix.

Also known as: delay, rtt, round-trip time

All glossary entries