active measurement

Imagine you want to know how busy a road is, so you send your own car down it with a stopwatch and time how long the trip takes. You are not watching the cars that happen to be there already — you are adding one of your own and learning from how it fares. Active measurement is exactly this idea for networks: you inject your own probe traffic and watch what the network does to it. The famous tools ping and traceroute are active measurements, and so is a speed test.

Concretely, an active tool generates packets on purpose, often with carefully chosen contents, and records what comes back and when. Ping sends a small probe and times how long until the reply returns. A bandwidth tester like iperf blasts data as fast as it can for a few seconds and reports the rate it achieved. A one-way delay measurement timestamps a packet when it leaves and again when it arrives. In every case the network is asked a question by being given traffic it would not otherwise have carried, and the answer is read from the timing, the losses, or the path the probes take.

The honest trade-off is that you are changing the very thing you measure. Probe traffic consumes bandwidth and buffer space, so a heavy active test can itself cause the congestion it is trying to detect — a kind of network Heisenberg problem. Active measurement is also only a sample: a ping every second tells you about those instants, not what happened between them. It is invaluable for asking a direct question on demand (is this host reachable? how fast is this link right now?), but for a continuous, low-impact picture of real user traffic you turn instead to passive measurement.