The complaint is always shaped the same way. The connection tests at full speed on a Tuesday afternoon. At nine in the evening the video call stutters, the stream drops to a lower bitrate, and a speed test returns a third of what you pay for. Nothing in the house changed. The cause is almost never your router — it is contention, and it is worth understanding properly, because it is the single largest determinant of whether a broadband plan is good.
What oversubscription actually is
No internet provider anywhere buys one unit of upstream capacity for every unit it sells. If a provider has ten thousand subscribers on 100 Mbps plans, provisioning a genuine terabit of transit for them would be ruinous and pointless: those ten thousand people are not all downloading at full tilt simultaneously, and never will be.
Instead, providers rely on statistical multiplexing. Traffic is bursty. Your connection is idle most of the time, busy for a few seconds while a page loads, then idle again. Averaged across thousands of subscribers, the aggregate demand is far smaller and far smoother than the sum of the headline speeds. Selling more capacity than you have bought is called oversubscription, and expressing it as a ratio — 20:1, 50:1 — gives you a contention ratio.
This is not, in itself, dishonest. It is how every shared network on earth works, including the mobile network your phone is on right now. The dishonesty, when it exists, is in the ratio: the difference between a provider that engineers for peak demand with headroom and one that keeps adding subscribers to a segment that is already saturated.
Where the bottleneck actually sits
“The internet is slow” can mean congestion at any of several points, and they behave differently.
- The access segment. On a GPON fibre tree, everyone behind one optical splitter shares roughly 2.5 Gbps down and 1.2 Gbps up. On a coaxial cable segment, everyone on that node shares the segment. If the split is dense and the plans are large, this saturates first.
- Aggregation and backhaul. The links carrying traffic from local nodes back to the provider’s core. Cheaper to under-build than the access layer, and invisible to the customer.
- Transit and peering. Where the provider hands traffic to the rest of the internet. A congested peering link with one particular content network produces the confusing symptom where one service is unwatchable and everything else is fine.
- Your own Wi-Fi. Genuinely common, and it looks identical from the sofa. Rule it out first — see how to test properly.
Why nine in the evening
Residential internet demand follows a sharply peaked daily curve. It is low overnight, moderate through the working day, and rises steeply from early evening as households return, eat, and settle in front of streaming video. Streaming is the dominant contributor: it is high-bitrate, sustained rather than bursty, and it is consumed at precisely the hours everyone else is consuming it. The statistical smoothing that makes oversubscription work at noon breaks down when a large fraction of subscribers want a steady 15–25 Mbps at the same moment.
The result is that a segment can be comfortable for twenty-two hours a day and congested for two. Which is why testing your connection at lunchtime tells you almost nothing about the product you bought.
How to tell whether contention is your problem
Contention has a recognisable signature. Four checks, in order:
- Test on a wire. Plug a computer directly into the router or ONT with an Ethernet cable. If the problem disappears, it was Wi-Fi, not the line.
- Test at three times of day. Mid-morning, late afternoon, and around 9–10 PM, on the same wired machine, to the same test server. Congestion produces a curve; a faulty line produces a flat low number all day.
- Watch latency, not just throughput. A congested path shows rising round-trip time and rising jitter before throughput collapses. If your ping to a nearby server goes from 8 ms at noon to 90 ms at nine, that is queueing on a saturated link.
- Test more than one destination. If a local test server is fast and everything international is slow, the problem is peering or transit rather than your local segment.
What Indian regulation says about this
TRAI has set a quality-of-service benchmark for broadband since its 2006 regulation on the subject: a subscriber should receive at least 80% of the subscribed speed, measured from the ISP node to the user. Those standards were consolidated and tightened in the Standards of Quality of Service of Access (Wireline and Wireless) and Broadband (Wireline and Wireless) Service Regulations, 2024, which came into force on 1 October 2024 and raised the penalty schedule for breaches.
Two honest caveats about relying on that number. First, the benchmark is measured to the provider’s node — it does not promise a particular speed to a server on another continent, and it cannot. Second, enforcement works on reported aggregates rather than on your individual line, so the benchmark is a floor for a provider’s behaviour, not a guarantee you can invoke on a bad evening. It is still a useful reference point when you raise a complaint, because it establishes that consistently receiving a third of your subscribed speed is not an acceptable outcome.
Questions worth asking before you sign
- What is the contention ratio on this plan, and is it stated anywhere in writing?
- Is capacity monitored per segment, and what utilisation threshold triggers an upgrade?
- Are there fair-usage or traffic-management rules that shape traffic during peak hours? “Unlimited” and “unshaped” are not the same claim.
- Can I get a wired speed test witnessed at installation, in the evening?
A provider that has genuinely engineered for peak will answer these readily; the policy is a selling point, not a secret. ValoNet’s position on this is a stated zero-oversubscription policy alongside its 99.9% uptime SLA — which is a claim you should hold any provider to rather than take on trust, and which is why the wording of an SLA repays close reading.