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

Goal

Produce a series of throughput and latency readings for one household connection that can be compared across days and time slots, with the variables that commonly move the number either held constant or written down.

Prerequisites

One client device kept for the whole series; a test tool that names its server and shows how many parallel connections it used; the ability to connect the device by cable to the router, or, failing that, a fixed position for it relative to the access point.

Steps

1. Fix the path and describe it in the header: device, operating system, cable or wireless, the access point and the device's distance and room if wireless, and the router model.
2. Fix the tool and the server. RFC 6349 derives the bandwidth-delay product from round-trip time and bottleneck bandwidth and states that it determines the socket buffer sizes required for maximum TCP throughput; its section on single versus multiple TCP connections explains that the choice depends on that product relative to the receive window. A tool that opens many connections can therefore report a different figure from a single-stream copy, so log the connection count the tool shows with every run.
3. Choose three daily slots (for example 07:30, 13:00, 21:00) and run three consecutive tests per slot for 14 days; log all nine readings per day, not the best.
4. Per run record: date, time, download, upload, idle latency, loaded latency if shown, server name, connection count, and household activity (number of devices active, backups or streaming running, other people at home).
5. Once a week, add a wired run in the same slot as a wireless run to separate the access network from the wireless link.
6. Weekly summary: per slot, the median, minimum and maximum of the download figures; keep the raw rows.

Expected result

A table in which a low reading can be attributed to a slot, a wireless path or a busy household before it is attributed to the connection, and in which a wired-versus-wireless gap is visible as a number.

Limits and test basis

Proposed protocol built on the cited RFC's framework, which states that it is intended for predictable managed IP networks with a service level agreement and that end users with best-effort access could use its methodology. A consumer test measures the path to one test server, uses the connection while measuring, and depends on the tool's parallelism; the log therefore describes this device, tool and path, and no claim about a provider or a contract follows from it.