Capacity and queues
Latency can rise before a link fails.
Every link and device has finite capacity. When traffic arrives faster than it can be transmitted or processed, buffers form a queue. The queue adds delay; if it fills, packets may be dropped and higher layers may retransmit them.
Congestion can occur at one hop, one access link, a wireless channel, a server, or an application. A fast test from one location does not rule out a busy path at another time.
Adapt the sending rate
Congestion signals can become feedback.
Senders can use rising delay, packet loss, or explicit network signals as evidence that they are offering traffic faster than the path can handle. They then reduce their sending rate, probe cautiously, and increase it again when conditions improve.
This feedback loop protects shared capacity, but it is not instantaneous. Several flows may react at different speeds, and an application can still overload a server even when the network path is healthy. Measure the sender, path, and receiver together.
Build the model
Four congestion signals
Queueing
Packets wait longer before a device can forward them.
Latency
Round-trip or application response time increases under load.
Loss
Full buffers or wireless contention can discard packets.
Recovery
Retransmissions and reduced sending rates change throughput.
Mini investigation
Inspect a congestion symptom
Select the observed behavior and identify the most direct next measurement.
Responses arrive, but waiting time increases while the path is busy.
Measure latency over time and compare idle and busy periods.
Knowledge check
Test your congestion model
Answer eight questions. The answer order changes each time.