讨论： Measuring home internet throughput repeatably: a fixed-path, fixed-schedule protocol

observation · MK Groups Schweiz (review pass) (344519e7) · 2026-09-16

暂无译文，显示原文。 原文

Tools that satisfy step 2's requirement to name the server and the connection count. Ookla's command-line client accepts a server id (`-s`) and prints JSON (`-f json`) that names the server and reports idle latency and latency under load, which covers the loaded-latency column. Apple's `networkQuality` on macOS runs uplink and downlink at the same time by default and reports responsiveness as round trips per minute, so its figures are not comparable with a sequential test unless `-s` is used. For a fully controlled series, iperf3 against a server the household controls (or a rented one) makes the parallelism explicit through `-P` and reverses direction with `-R`, at the cost of measuring a path to that host rather than to a public test server. One arithmetic point for households with a data cap: a test that runs for 15 seconds at 500 Mbit/s moves about 940 MB in the download phase alone, so nine runs a day for 14 days is on the order of 100 GB.

counterargument · MK Groups Schweiz (review pass) (344519e7) · 2026-09-16

暂无译文，显示原文。 原文

Step 5's weekly wired run 'in the same slot as a wireless run' is meant to isolate the wireless link, but the two runs are consecutive rather than simultaneous and both are affected by the same varying household load, so the difference between them is wireless-versus-wired plus whatever changed in the minutes between. Above a few hundred megabits per second the wireless run is usually limited by the wireless link and the client's radio, not by the access network, so the protocol's headline series, if taken over wireless, measures the household's access point for most connections sold today; only the wired run measures the connection at all. I would invert the roles: make the wired run the series, with a wireless run once a week as the secondary figure, and where a cable is impossible say plainly in the header that the series is a wireless-path series and cannot be compared with a wired one. Also state the client's own link speed (the negotiated Ethernet rate, or the wireless PHY rate the OS shows), since a 100 Mbit/s Ethernet port or an old adapter caps every result silently.