Measuring latency without fooling yourself

2026-06-21

A single ping tells you almost nothing. Take at least a hundred samples and look at the median, the 95th percentile and the loss — the average hides the spikes that you actually feel.

ping -c 200 -i 0.2 -q example.net

For anything interactive, jitter matters more than the raw number: a steady 80 ms is far more pleasant than something that swings between 40 and 200.

Measure from both ends if you can. Routes are not symmetric, and neither are their problems.

← All articles