Telemetry — Network
What the internet costs, depending on where I sit
Every few minutes my laptop times how long it takes a small packet to reach a few well-known servers and come back. Over 189 days it did this from 8 different kinds of place — home, a relative's house, hotels, libraries, two airports, and a plane.
The gap between best and worst is 69×, and the ranking is not the one you'd guess.
Readings through rolled up page built
Airports are fast. Aeroplanes are not.
Each row is a kind of place. The filled dot is the typical round trip; the hollow dot is a bad-but-not-unusual one, the sort of delay you hit about one time in ten. The scale is logarithmic, because otherwise a single row flattens all the others into a smear.
Public Wi-Fi has a bad reputation it doesn't deserve here: airport and library connections came in at 10 milliseconds, comfortably quicker than my own house at 24. Home is steadier — its best and worst are close together — but it is not faster.
And then there is in-flight Wi-Fi, at 660 milliseconds. That is not a bad connection, it is a different category of thing: the signal is going up to a satellite and back. Two thirds of a second before anything you click begins to happen.
Some of these rest on very few readings — I was in an airport for hours, not months. The counts are shown so you can weigh them accordingly. Network names never leave the processing step: each becomes a category like “Home” or “Airport” and the original is discarded.
What a connection usually feels like
The usual way to show this is a line over time, which answers what happened on Tuesday. These are distributions instead: how often each response time occurred, across every reading ever taken. Taller bars on the left mean a snappier connection.
The gateway — the router in the next room — is the leftmost and tightest, as it should be. The three public servers sit further right and spread wider. All four have a long thin tail on the right: about 0.20% of readings land in the slowest bins, mostly the moment after it wakes from sleep and the network hasn't reconnected yet.
Those few extreme readings are exactly why a distribution beats a timeline. On a line chart a handful of multi-second spikes stretch the scale until every normal day is a flat line at the bottom. Here they simply occupy the last bin and stop distorting everything else.
Horizontal axis is round-trip time in milliseconds, on bins that widen as they go right.
1,963 GB down, 1,075 GB up
Data pulled down each day. It averages 10.4 GB, with a peak of 36.6 GB on 2026-07-31. For one machine and one person over six months that is a lot. It is what moving datasets around for a living looks like from outside.
The upload figure is the one I find more telling. Most consumer internet use is lopsided, with downloads dwarfing uploads. Here it's roughly two to one, which is the signature of pushing things to servers rather than watching things.