Latency

Latency is the time between firing a request and receiving the start of the response, measured in milliseconds. It is the delay — not the amount of data that fits along the way.

In online games and home internet, the same delay is often called ping: it is the same quantity, measured from a different point and with a different commercial name.

Latency is not connection speed

Bandwidth is the width of the pipe; latency is how long the first drop takes to arrive. A 500 Mbps link with 300 ms of latency delivers a lot of data and feels slow, because every round trip costs three tenths of a second. That is why satellite internet downloads a movie fast and stutters on a voice call.

Where the delay comes from

• Physical distance. Light in fiber travels at about 200,000 km/s. A packet from São Paulo to the US East Coast and back rarely stays under 120 ms, even on an empty network. It is physics, not configuration.
• Processing. A heavy database query, an AI-generated reply, a call to a third party along the way.
• A queue. The server is busy and your request waits its turn — this is what fires during a campaign spike.

The average lies

An average hides the irritated customer. If 95 requests take 100 ms and five take 6 seconds, the average sits at 395 ms and looks acceptable — but five out of every hundred people waited six seconds. Use p95 and p99: the time 95% and 99% of calls did not exceed. It is the only measure that sees the tail.

What is acceptable

In support, system latency and response time to the customer are quantities three orders of magnitude apart: the first is measured in milliseconds, the second in minutes. Optimizing milliseconds while a lead waits forty minutes is solving the wrong problem.