Contention at Teatime: A Month of Full-Fibre Throughput, Latency and Loss in a UK Home

Full-fibre marketing has a habit of arriving as a single number: 900 Mbps, 1 Gbps, 1.6 Gbps. The number is useful, but it is not a description of an evening. Between roughly 19:00 and 22:00, a UK street can change from a collection of idle lines into a shared backhaul problem, with video calls, game updates, streaming and cloud backups all competing for the same capacity. To see how much of the advertised rate survives that transition, CavLab instrumented a residential FTTP connection in a medium-sized English town for 30 days in autumn 2026. The service was a 900 Mbps symmetrical package sold on a full-fibre access network.

What we measured, and what we deliberately did not

The test rig was deliberately boring. A small fanless PC connected by Ethernet to the ISP-supplied router ran scheduled throughput tests against three nearby HTTP speed-test endpoints and an iperf3 target in a London cloud region. Every 15 minutes it recorded download and upload throughput, idle latency to the first IP hop and to a London anycast address, loaded latency during a saturating transfer, and packet loss over 60-second windows. A separate Wi-Fi 6 laptop ran the same tests at two positions: two metres from the router with line of sight, and ten metres away through two internal walls. We defined peak as 19:00-22:00 on weekdays and off-peak as 02:00-05:00. No test was run while the household was deliberately streaming or gaming, so the figures describe the line rather than a worst-case household.

This matters because most broadband arguments confuse three different things: the access line, the home network and the server at the other end. A speed test is a chain of all three. If the Wi-Fi is poor, the line can be perfect and the result will still be disappointing.

The headline rate is a median, not a promise at every socket

The package advertises 900 Mbps downstream and 900 Mbps upstream. Over the month, wired downloads at peak had a median of 871 Mbps, with an upstream median of 858 Mbps. Off-peak, the same wired client recorded 918 Mbps down and 902 Mbps up. In other words, the line delivered about 97% of the advertised download rate at the median during the evening and slightly more than the headline figure overnight, when overheads and test-server variance allow it.

The distribution is more interesting than the median. The tenth percentile of peak wired downloads was 612 Mbps. On two evenings, a sustained test fell below 500 Mbps for more than ten minutes. These were not router reboots or line drops; latency rose at the same time, which points to contention upstream rather than a local fault. A single speed test at 20:30 could therefore show a very good number or a merely adequate one, depending on the minute.

Wired Ethernet mostly keeps its side of the bargain

For anyone who can use a cable, the result is reassuring. Wired latency was 9 ms median at peak and 7 ms off-peak to the London target. Loaded latency, measured while the line was saturated, rose to 38 ms at peak and 22 ms off-peak. Packet loss stayed below 0.06% in peak windows and below 0.01% overnight. Those are not lab-clean figures, but they are comfortably good enough for video conferencing, remote desktop work and competitive gaming, provided the home router is not adding its own queueing delay.

The caveat is the router. The ISP-supplied unit handled 900 Mbps in our wired tests, but its Wi-Fi and bufferbloat behaviour were unremarkable. A separate router with active queue management would probably have reduced loaded latency further. That is an in-home upgrade, not a line upgrade.

Wi-Fi is where the evening story changes

At two metres with line of sight, the Wi-Fi 6 laptop reached a median of 690 Mbps down and 640 Mbps up at peak. Ten metres away, through two walls, the same client managed 210 Mbps down and 180 Mbps up. Off-peak the near-position figure rose to 760 Mbps; the far position improved only modestly. The lesson is not that Wi-Fi is bad. It is that a 900 Mbps service is a wired specification, and the wireless link is a shared, variable and interference-prone segment that the ISP does not control.

That matters for the common UK layout: router in the hall, living room through one wall, bedroom through two. A 4K stream at 25 Mbps is fine. A 4K stream, a game download, a video call and a phone backup at the same time is where the far-room client starts to queue. The access line still has headroom; the home network does not.

Latency, loss and the 8pm cliff

The clearest peak-time signal was not throughput but latency. Idle latency rose by roughly 2 ms in the evening, and loaded latency rose by 16 ms. Packet loss remained low but became more variable, with short bursts during the two sub-500 Mbps events. For work-from-home users, that pattern is more noticeable than a drop from 900 to 700 Mbps. Teams and Zoom adapt to bandwidth; they cope less gracefully with jitter and loss. For gamers, the same 9 ms idle latency is good, but a router that lets a large download inflate latency to 100 ms will feel worse than a slower line with better queue management.

Streaming is the least sensitive workload. Netflix, iPlayer and Disney+ recommend well under 50 Mbps for 4K, and adaptive bitrate systems handle short dips. The evening experience for a single stream is unlikely to differ between a 500 Mbps and a 900 Mbps line. The difference appears when several people are active, or when a console decides to download a 100 GB update at 19:30.

Reading the marketing against the Ofcom code

The Ofcom broadband speeds code of practice asks ISPs to give customers a clear estimate of the speed they can expect at peak times, and to provide a route out if the service consistently falls short. It does not turn a 900 Mbps headline into a guarantee of 900 Mbps on every device at every hour. Advertising rules also expect headline fixed-line speeds to be based on median peak-time performance rather than a best-case lab result. Our wired median of 871 Mbps is consistent with that idea. The tenth-percentile figure is the number a careful buyer should ask about, because it describes the evenings when the street is busiest.

The practical test for a reader is simple: run a wired speed test at 20:00 and again at 03:00 for a week. If the wired peak median is close to the advertised rate and the off-peak figure is higher, the line is behaving normally. If the wired peak figure is consistently far below the estimate, document it and approach the ISP. If only Wi-Fi is slow, fix the home network first.

What the month showed

Full fibre in this test did what full fibre should do: it delivered close to the advertised rate on Ethernet, with low latency and negligible loss for most of the month. The evening did not break it. It introduced contention, visible in the tail of the throughput distribution and in loaded latency. The real 8pm cliff was mostly wireless and mostly local. A gigabit line is a strong foundation for working from home, gaming and streaming, but it is not a substitute for a decent router, a wired connection where it matters, and a realistic reading of what a median speed claim actually means.