4 November 2026 · Networking · 7 min read

Unmetered vs Metered Bandwidth: What a Server Port Actually Costs

Unmetered vs Metered Bandwidth: What a Server Port Actually Costs

Every hosting quote has a bandwidth line, and it's the line most buyers misread. "Unmetered 1 Gbps" sounds like infinity. "10 TB metered" sounds like a limit. Neither reading survives contact with how traffic is actually measured and billed. This is a plain walkthrough of the two models, the arithmetic behind 95th percentile billing, and how to work out which one your workload wants.

What unmetered actually means

Unmetered is a port cap, not an all-you-can-eat promise. An unmetered 1 Gbps port means the provider doesn't count your bytes; you can push as much as fits down a 1 Gbps pipe. The ceiling is physical. A gigabit port that runs flat out, 24 hours a day, for a month can move roughly 324 TB. Nobody's workload does that, but the number is useful as a scale reference: an unmetered 1 Gbps service is a promise about rate, not volume.

That distinction matters in both directions. A file server that bursts hard for two hours a night fits an unmetered port well. A video streaming service that averages 700 Mbps around the clock will hit the port cap constantly, and an unmetered 100 Mbps or 1 Gbps plan is the wrong shape of product entirely, because your traffic is rate-limited at exactly the point it wants to run.

How 95th percentile billing works

Metered plans come in two sub-flavours, and the difference is where the trap sits. Fixed-volume plans, "10 TB per month, overage at X per TB", are simple and fine for predictable traffic. The one to read carefully is burstable 95th percentile billing, common on colocation and higher-end dedicated quotes.

The mechanics: the switch samples your port every five minutes, about 8,640 samples a month. Each sample is the average rate over that five minutes. The provider sorts all samples from highest to lowest, throws away the top 5 percent, and bills you at the rate of the highest sample remaining.

The 36 free samples are the whole game. You can burst above your committed rate for up to 36 five-minute windows a month, about three hours in total, at no charge. A nightly backup that saturates the port for 20 minutes a night stays under the free window. One burst that runs four hours pushes you over, and you pay the raised rate for the whole month, because the 95th percentile is a monthly figure, not an average. Read a burstable quote and ask: what's my commit, what's the per-Mbps overage, and is overage billed at commit rate or a higher burst rate?

A worked example

Say a server averages 120 Mbps during UK business hours, drops to near nothing overnight, and runs a 30-minute backup burst at 900 Mbps each night.

On 95th percentile billing, the overnight bursts alone are about 30 windows a night times 22 working nights, far beyond 36 samples, so the burst sets your bill. The 95th percentile lands around 900 Mbps and you pay for a rate your average traffic never approaches. Your actual monthly volume is modest: about 6 TB. A fixed 10 TB metered plan would carry it easily and cheaply.

Flip the workload. A busy API server holding a steady 300 Mbps for 10 hours a day has a 95th percentile near 300 Mbps but a monthly volume near 40 TB. On volume pricing that's expensive; on an unmetered 1 Gbps port it's comfortable and the bill never moves. Same numbers, opposite conclusions. The shape of the traffic decides, not the total.

The hidden lines in the quote

Which one to pick

Working rules from the racks we run at ServerCabin:

The bandwidth line is also where hosting marketing does most of its lounging. A quote that leads with "unlimited" and doesn't state the port speed, or states a volume without stating whether overage is blocked or billed, isn't telling you the price. The full list of lines worth demanding on a hosting proposal is in our SLA writeup, and the cost crossover between renting iron and racking your own, where bandwidth pricing decides half the answer, is in our colocation vs dedicated comparison.

Measure first. One week of five-minute samples from your current server, sorted, tells you your real 95th percentile and your real volume, and the right billing model is obvious from there. Mail admin@servercabin.com with the numbers and we'll tell you which shape fits, including when the honest answer is the cheaper plan.